)]}'
{"doc/admin-guide/source/compute-enhanced-platform-awareness.rst":[{"author":{"_account_id":14962,"name":"ologvinova","email":"ologvinova@mirantis.com","username":"ologvinova"},"change_message_id":"1cdd9c80f23b6c6e5f9852adc8175e3929f73340","unresolved":false,"context_lines":[{"line_number":1,"context_line":"\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d"},{"line_number":2,"context_line":"Enhanced Platform Awareness"},{"line_number":3,"context_line":"\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d"},{"line_number":4,"context_line":""},{"line_number":5,"context_line":"OpenStack clouds run on platforms that differ greatly in the capabilities that"}],"source_content_type":"text/x-rst","patch_set":1,"id":"1a122d0e_d0b60188","line":2,"range":{"start_line":2,"start_character":9,"end_line":2,"end_character":10},"updated":"2016-05-05 15:27:58.000000000","message":"Could you please use sentence-case capitalization for the headings in your patch? -\u003e Enhanced platform awareness etc.","commit_id":"78f32244badbaef989e826bf02fac2c14b7a6041"},{"author":{"_account_id":15334,"name":"Stephen Finucane","display_name":"stephenfin","email":"stephenfin@redhat.com","username":"sfinucan"},"change_message_id":"de849e581797cb26a10c9897f57d162af984b387","unresolved":false,"context_lines":[{"line_number":1,"context_line":"\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d"},{"line_number":2,"context_line":"Enhanced Platform Awareness"},{"line_number":3,"context_line":"\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d"},{"line_number":4,"context_line":""},{"line_number":5,"context_line":"OpenStack clouds run on platforms that differ greatly in the capabilities that"}],"source_content_type":"text/x-rst","patch_set":1,"id":"dab17558_3920a19b","line":2,"range":{"start_line":2,"start_character":9,"end_line":2,"end_character":10},"in_reply_to":"1a122d0e_d0b60188","updated":"2016-05-06 10:05:10.000000000","message":"Done","commit_id":"78f32244badbaef989e826bf02fac2c14b7a6041"},{"author":{"_account_id":17973,"name":"Ludovic Beliveau","email":"lbeliveau@gmail.com","username":"lbeliveau"},"change_message_id":"8968f9ec825172062aee51bc0ea57f4f323df113","unresolved":false,"context_lines":[{"line_number":6,"context_line":"they provide. By default, the Compute service seeks to abstract the underlying"},{"line_number":7,"context_line":"hardware that it runs on, rather than exposing specifics about the underlying"},{"line_number":8,"context_line":"host platforms. This abstraction manifests itself in many ways. For example,"},{"line_number":9,"context_line":"rather than exposing the types and toplogies of CPUs running on hosts, the"},{"line_number":10,"context_line":"service exposes a number of generic CPUs (virtual CPUs, or vCPUs) and allows"},{"line_number":11,"context_line":"for overprovisioning of these. In a similar manner, rather than exposing the"},{"line_number":12,"context_line":"individual types of network devices available on hosts, generic"}],"source_content_type":"text/x-rst","patch_set":1,"id":"1a122d0e_293a70cd","line":9,"range":{"start_line":9,"start_character":36,"end_line":9,"end_character":37},"updated":"2016-05-05 16:12:54.000000000","message":"s/toplogies/topologies/","commit_id":"78f32244badbaef989e826bf02fac2c14b7a6041"},{"author":{"_account_id":15334,"name":"Stephen Finucane","display_name":"stephenfin","email":"stephenfin@redhat.com","username":"sfinucan"},"change_message_id":"de849e581797cb26a10c9897f57d162af984b387","unresolved":false,"context_lines":[{"line_number":6,"context_line":"they provide. By default, the Compute service seeks to abstract the underlying"},{"line_number":7,"context_line":"hardware that it runs on, rather than exposing specifics about the underlying"},{"line_number":8,"context_line":"host platforms. This abstraction manifests itself in many ways. For example,"},{"line_number":9,"context_line":"rather than exposing the types and toplogies of CPUs running on hosts, the"},{"line_number":10,"context_line":"service exposes a number of generic CPUs (virtual CPUs, or vCPUs) and allows"},{"line_number":11,"context_line":"for overprovisioning of these. In a similar manner, rather than exposing the"},{"line_number":12,"context_line":"individual types of network devices available on hosts, generic"}],"source_content_type":"text/x-rst","patch_set":1,"id":"dab17558_195cbd24","line":9,"range":{"start_line":9,"start_character":36,"end_line":9,"end_character":37},"in_reply_to":"1a122d0e_293a70cd","updated":"2016-05-06 10:05:10.000000000","message":"Done","commit_id":"78f32244badbaef989e826bf02fac2c14b7a6041"},{"author":{"_account_id":17973,"name":"Ludovic Beliveau","email":"lbeliveau@gmail.com","username":"lbeliveau"},"change_message_id":"8968f9ec825172062aee51bc0ea57f4f323df113","unresolved":false,"context_lines":[{"line_number":8,"context_line":"host platforms. This abstraction manifests itself in many ways. For example,"},{"line_number":9,"context_line":"rather than exposing the types and toplogies of CPUs running on hosts, the"},{"line_number":10,"context_line":"service exposes a number of generic CPUs (virtual CPUs, or vCPUs) and allows"},{"line_number":11,"context_line":"for overprovisioning of these. In a similar manner, rather than exposing the"},{"line_number":12,"context_line":"individual types of network devices available on hosts, generic"},{"line_number":13,"context_line":"software-powered network ports are provided. These features are designed to"},{"line_number":14,"context_line":"enable high resource utilization, and allows the service to provide a generic,"}],"source_content_type":"text/x-rst","patch_set":1,"id":"1a122d0e_a93680b1","line":11,"updated":"2016-05-05 16:12:54.000000000","message":"s/overprovisioning/over-provisioning/\n\nNot sure I understand overprovisionning in the case that interests us.  The scheduler won\u0027t be able to spawn guess if it\u0027s over-provisionned right ?  I\u0027m not too familiar with pinning, it might just need some more clarification.","commit_id":"78f32244badbaef989e826bf02fac2c14b7a6041"},{"author":{"_account_id":15334,"name":"Stephen Finucane","display_name":"stephenfin","email":"stephenfin@redhat.com","username":"sfinucan"},"change_message_id":"de849e581797cb26a10c9897f57d162af984b387","unresolved":false,"context_lines":[{"line_number":8,"context_line":"host platforms. This abstraction manifests itself in many ways. For example,"},{"line_number":9,"context_line":"rather than exposing the types and toplogies of CPUs running on hosts, the"},{"line_number":10,"context_line":"service exposes a number of generic CPUs (virtual CPUs, or vCPUs) and allows"},{"line_number":11,"context_line":"for overprovisioning of these. In a similar manner, rather than exposing the"},{"line_number":12,"context_line":"individual types of network devices available on hosts, generic"},{"line_number":13,"context_line":"software-powered network ports are provided. These features are designed to"},{"line_number":14,"context_line":"enable high resource utilization, and allows the service to provide a generic,"}],"source_content_type":"text/x-rst","patch_set":1,"id":"dab17558_79c4e991","line":11,"in_reply_to":"1a122d0e_a93680b1","updated":"2016-05-06 10:05:10.000000000","message":"So in this paragraph I\u0027m trying to indicate how the compute service exposes resource _without_ features like pinning etc. When pinning is enabled then over-provisioning is no longer available, as you say. I\u0027ll attempt to clarify this","commit_id":"78f32244badbaef989e826bf02fac2c14b7a6041"},{"author":{"_account_id":14962,"name":"ologvinova","email":"ologvinova@mirantis.com","username":"ologvinova"},"change_message_id":"1cdd9c80f23b6c6e5f9852adc8175e3929f73340","unresolved":false,"context_lines":[{"line_number":11,"context_line":"for overprovisioning of these. In a similar manner, rather than exposing the"},{"line_number":12,"context_line":"individual types of network devices available on hosts, generic"},{"line_number":13,"context_line":"software-powered network ports are provided. These features are designed to"},{"line_number":14,"context_line":"enable high resource utilization, and allows the service to provide a generic,"},{"line_number":15,"context_line":"cost-effective and highly-scalable cloud upon which to build applications."},{"line_number":16,"context_line":""},{"line_number":17,"context_line":"This abstraction is beneficial for most applications. However, there are some"}],"source_content_type":"text/x-rst","patch_set":1,"id":"1a122d0e_d36e0bc7","line":14,"range":{"start_line":14,"start_character":21,"end_line":14,"end_character":44},"updated":"2016-05-05 15:27:58.000000000","message":"-\u003e utilization and allow (the features allow, right?)","commit_id":"78f32244badbaef989e826bf02fac2c14b7a6041"},{"author":{"_account_id":14962,"name":"ologvinova","email":"ologvinova@mirantis.com","username":"ologvinova"},"change_message_id":"1cdd9c80f23b6c6e5f9852adc8175e3929f73340","unresolved":false,"context_lines":[{"line_number":11,"context_line":"for overprovisioning of these. In a similar manner, rather than exposing the"},{"line_number":12,"context_line":"individual types of network devices available on hosts, generic"},{"line_number":13,"context_line":"software-powered network ports are provided. These features are designed to"},{"line_number":14,"context_line":"enable high resource utilization, and allows the service to provide a generic,"},{"line_number":15,"context_line":"cost-effective and highly-scalable cloud upon which to build applications."},{"line_number":16,"context_line":""},{"line_number":17,"context_line":"This abstraction is beneficial for most applications. However, there are some"}],"source_content_type":"text/x-rst","patch_set":1,"id":"1a122d0e_13c1a380","line":14,"range":{"start_line":14,"start_character":77,"end_line":14,"end_character":78},"updated":"2016-05-05 15:27:58.000000000","message":"I would remove a comma here.","commit_id":"78f32244badbaef989e826bf02fac2c14b7a6041"},{"author":{"_account_id":15334,"name":"Stephen Finucane","display_name":"stephenfin","email":"stephenfin@redhat.com","username":"sfinucan"},"change_message_id":"de849e581797cb26a10c9897f57d162af984b387","unresolved":false,"context_lines":[{"line_number":11,"context_line":"for overprovisioning of these. In a similar manner, rather than exposing the"},{"line_number":12,"context_line":"individual types of network devices available on hosts, generic"},{"line_number":13,"context_line":"software-powered network ports are provided. These features are designed to"},{"line_number":14,"context_line":"enable high resource utilization, and allows the service to provide a generic,"},{"line_number":15,"context_line":"cost-effective and highly-scalable cloud upon which to build applications."},{"line_number":16,"context_line":""},{"line_number":17,"context_line":"This abstraction is beneficial for most applications. However, there are some"}],"source_content_type":"text/x-rst","patch_set":1,"id":"dab17558_f95d79ed","line":14,"range":{"start_line":14,"start_character":77,"end_line":14,"end_character":78},"in_reply_to":"1a122d0e_13c1a380","updated":"2016-05-06 10:05:10.000000000","message":"I\u0027m not so sure about this one (it\u0027s a list), but it\u0027s done anyway","commit_id":"78f32244badbaef989e826bf02fac2c14b7a6041"},{"author":{"_account_id":15334,"name":"Stephen Finucane","display_name":"stephenfin","email":"stephenfin@redhat.com","username":"sfinucan"},"change_message_id":"de849e581797cb26a10c9897f57d162af984b387","unresolved":false,"context_lines":[{"line_number":11,"context_line":"for overprovisioning of these. In a similar manner, rather than exposing the"},{"line_number":12,"context_line":"individual types of network devices available on hosts, generic"},{"line_number":13,"context_line":"software-powered network ports are provided. These features are designed to"},{"line_number":14,"context_line":"enable high resource utilization, and allows the service to provide a generic,"},{"line_number":15,"context_line":"cost-effective and highly-scalable cloud upon which to build applications."},{"line_number":16,"context_line":""},{"line_number":17,"context_line":"This abstraction is beneficial for most applications. However, there are some"}],"source_content_type":"text/x-rst","patch_set":1,"id":"dab17558_195ddde9","line":14,"range":{"start_line":14,"start_character":21,"end_line":14,"end_character":44},"in_reply_to":"1a122d0e_d36e0bc7","updated":"2016-05-06 10:05:10.000000000","message":"Done","commit_id":"78f32244badbaef989e826bf02fac2c14b7a6041"},{"author":{"_account_id":14962,"name":"ologvinova","email":"ologvinova@mirantis.com","username":"ologvinova"},"change_message_id":"1cdd9c80f23b6c6e5f9852adc8175e3929f73340","unresolved":false,"context_lines":[{"line_number":31,"context_line":"Non-uniform memory access (NUMA) is a design found in many modern,"},{"line_number":32,"context_line":"multi-socket systems. In a NUMA system, system memory is divided into cells"},{"line_number":33,"context_line":"or nodes that are associated with particular CPUs. Requests for memory on"},{"line_number":34,"context_line":"other nodess is possible through an interconnect bus. However, bandwidth"},{"line_number":35,"context_line":"across this bus is shared and limited."},{"line_number":36,"context_line":""},{"line_number":37,"context_line":"When running workloads on NUMA hosts, it is important that the vCPUs executing"}],"source_content_type":"text/x-rst","patch_set":1,"id":"1a122d0e_136fe392","line":34,"range":{"start_line":34,"start_character":13,"end_line":34,"end_character":15},"updated":"2016-05-05 15:27:58.000000000","message":"are (requests are)","commit_id":"78f32244badbaef989e826bf02fac2c14b7a6041"},{"author":{"_account_id":14962,"name":"ologvinova","email":"ologvinova@mirantis.com","username":"ologvinova"},"change_message_id":"1cdd9c80f23b6c6e5f9852adc8175e3929f73340","unresolved":false,"context_lines":[{"line_number":31,"context_line":"Non-uniform memory access (NUMA) is a design found in many modern,"},{"line_number":32,"context_line":"multi-socket systems. In a NUMA system, system memory is divided into cells"},{"line_number":33,"context_line":"or nodes that are associated with particular CPUs. Requests for memory on"},{"line_number":34,"context_line":"other nodess is possible through an interconnect bus. However, bandwidth"},{"line_number":35,"context_line":"across this bus is shared and limited."},{"line_number":36,"context_line":""},{"line_number":37,"context_line":"When running workloads on NUMA hosts, it is important that the vCPUs executing"}],"source_content_type":"text/x-rst","patch_set":1,"id":"1a122d0e_f35347dc","line":34,"range":{"start_line":34,"start_character":6,"end_line":34,"end_character":12},"updated":"2016-05-05 15:27:58.000000000","message":"nodes","commit_id":"78f32244badbaef989e826bf02fac2c14b7a6041"},{"author":{"_account_id":15334,"name":"Stephen Finucane","display_name":"stephenfin","email":"stephenfin@redhat.com","username":"sfinucan"},"change_message_id":"de849e581797cb26a10c9897f57d162af984b387","unresolved":false,"context_lines":[{"line_number":31,"context_line":"Non-uniform memory access (NUMA) is a design found in many modern,"},{"line_number":32,"context_line":"multi-socket systems. In a NUMA system, system memory is divided into cells"},{"line_number":33,"context_line":"or nodes that are associated with particular CPUs. Requests for memory on"},{"line_number":34,"context_line":"other nodess is possible through an interconnect bus. However, bandwidth"},{"line_number":35,"context_line":"across this bus is shared and limited."},{"line_number":36,"context_line":""},{"line_number":37,"context_line":"When running workloads on NUMA hosts, it is important that the vCPUs executing"}],"source_content_type":"text/x-rst","patch_set":1,"id":"dab17558_d90995df","line":34,"range":{"start_line":34,"start_character":13,"end_line":34,"end_character":15},"in_reply_to":"1a122d0e_136fe392","updated":"2016-05-06 10:05:10.000000000","message":"Done","commit_id":"78f32244badbaef989e826bf02fac2c14b7a6041"},{"author":{"_account_id":15334,"name":"Stephen Finucane","display_name":"stephenfin","email":"stephenfin@redhat.com","username":"sfinucan"},"change_message_id":"de849e581797cb26a10c9897f57d162af984b387","unresolved":false,"context_lines":[{"line_number":31,"context_line":"Non-uniform memory access (NUMA) is a design found in many modern,"},{"line_number":32,"context_line":"multi-socket systems. In a NUMA system, system memory is divided into cells"},{"line_number":33,"context_line":"or nodes that are associated with particular CPUs. Requests for memory on"},{"line_number":34,"context_line":"other nodess is possible through an interconnect bus. However, bandwidth"},{"line_number":35,"context_line":"across this bus is shared and limited."},{"line_number":36,"context_line":""},{"line_number":37,"context_line":"When running workloads on NUMA hosts, it is important that the vCPUs executing"}],"source_content_type":"text/x-rst","patch_set":1,"id":"dab17558_390f01e3","line":34,"range":{"start_line":34,"start_character":6,"end_line":34,"end_character":12},"in_reply_to":"1a122d0e_f35347dc","updated":"2016-05-06 10:05:10.000000000","message":"Done","commit_id":"78f32244badbaef989e826bf02fac2c14b7a6041"},{"author":{"_account_id":14962,"name":"ologvinova","email":"ologvinova@mirantis.com","username":"ologvinova"},"change_message_id":"1cdd9c80f23b6c6e5f9852adc8175e3929f73340","unresolved":false,"context_lines":[{"line_number":39,"context_line":"ensures all memory accesses are local to the NUMA node and thus do not consume"},{"line_number":40,"context_line":"the limited cross-node memory bandwidth, adding latency to memory accesses. In"},{"line_number":41,"context_line":"addition, hugepages are assigned from memory and benefit from the same"},{"line_number":42,"context_line":"performance improvements as memory allocated using standard pages, thus they"},{"line_number":43,"context_line":"should be local also. Finally, PCI devices are directly associated with"},{"line_number":44,"context_line":"specific NUMA nodes for the purposes of DMA, so when using PCI device"},{"line_number":45,"context_line":"assignment it is also desirable that the guest be placed on the same NUMA node"}],"source_content_type":"text/x-rst","patch_set":1,"id":"1a122d0e_d3abab86","line":42,"range":{"start_line":42,"start_character":71,"end_line":42,"end_character":72},"updated":"2016-05-05 15:27:58.000000000","message":"comma","commit_id":"78f32244badbaef989e826bf02fac2c14b7a6041"},{"author":{"_account_id":15334,"name":"Stephen Finucane","display_name":"stephenfin","email":"stephenfin@redhat.com","username":"sfinucan"},"change_message_id":"de849e581797cb26a10c9897f57d162af984b387","unresolved":false,"context_lines":[{"line_number":39,"context_line":"ensures all memory accesses are local to the NUMA node and thus do not consume"},{"line_number":40,"context_line":"the limited cross-node memory bandwidth, adding latency to memory accesses. In"},{"line_number":41,"context_line":"addition, hugepages are assigned from memory and benefit from the same"},{"line_number":42,"context_line":"performance improvements as memory allocated using standard pages, thus they"},{"line_number":43,"context_line":"should be local also. Finally, PCI devices are directly associated with"},{"line_number":44,"context_line":"specific NUMA nodes for the purposes of DMA, so when using PCI device"},{"line_number":45,"context_line":"assignment it is also desirable that the guest be placed on the same NUMA node"}],"source_content_type":"text/x-rst","patch_set":1,"id":"dab17558_d9d7d521","line":42,"range":{"start_line":42,"start_character":71,"end_line":42,"end_character":72},"in_reply_to":"1a122d0e_d3abab86","updated":"2016-05-06 10:05:10.000000000","message":"Done","commit_id":"78f32244badbaef989e826bf02fac2c14b7a6041"},{"author":{"_account_id":19779,"name":"Chason Chan","display_name":"chenxing","email":"chason.chan@foxmail.com","username":"chenxing"},"change_message_id":"cb2d23d337d283cf9c50987eb0bd6446ce424525","unresolved":false,"context_lines":[{"line_number":39,"context_line":"ensures all memory accesses are local to the NUMA node and thus do not consume"},{"line_number":40,"context_line":"the limited cross-node memory bandwidth, adding latency to memory accesses. In"},{"line_number":41,"context_line":"addition, hugepages are assigned from memory and benefit from the same"},{"line_number":42,"context_line":"performance improvements as memory allocated using standard pages, thus they"},{"line_number":43,"context_line":"should be local also. Finally, PCI devices are directly associated with"},{"line_number":44,"context_line":"specific NUMA nodes for the purposes of DMA, so when using PCI device"},{"line_number":45,"context_line":"assignment it is also desirable that the guest be placed on the same NUMA node"},{"line_number":46,"context_line":"as any PCI device that is assigned to it."}],"source_content_type":"text/x-rst","patch_set":1,"id":"1a122d0e_a4786790","line":43,"range":{"start_line":42,"start_character":67,"end_line":43,"end_character":20},"updated":"2016-05-05 15:56:20.000000000","message":"they also should...","commit_id":"78f32244badbaef989e826bf02fac2c14b7a6041"},{"author":{"_account_id":15334,"name":"Stephen Finucane","display_name":"stephenfin","email":"stephenfin@redhat.com","username":"sfinucan"},"change_message_id":"de849e581797cb26a10c9897f57d162af984b387","unresolved":false,"context_lines":[{"line_number":39,"context_line":"ensures all memory accesses are local to the NUMA node and thus do not consume"},{"line_number":40,"context_line":"the limited cross-node memory bandwidth, adding latency to memory accesses. In"},{"line_number":41,"context_line":"addition, hugepages are assigned from memory and benefit from the same"},{"line_number":42,"context_line":"performance improvements as memory allocated using standard pages, thus they"},{"line_number":43,"context_line":"should be local also. Finally, PCI devices are directly associated with"},{"line_number":44,"context_line":"specific NUMA nodes for the purposes of DMA, so when using PCI device"},{"line_number":45,"context_line":"assignment it is also desirable that the guest be placed on the same NUMA node"},{"line_number":46,"context_line":"as any PCI device that is assigned to it."}],"source_content_type":"text/x-rst","patch_set":1,"id":"dab17558_7995093f","line":43,"range":{"start_line":42,"start_character":67,"end_line":43,"end_character":20},"in_reply_to":"1a122d0e_a4786790","updated":"2016-05-06 10:05:10.000000000","message":"Done","commit_id":"78f32244badbaef989e826bf02fac2c14b7a6041"},{"author":{"_account_id":17973,"name":"Ludovic Beliveau","email":"lbeliveau@gmail.com","username":"lbeliveau"},"change_message_id":"8968f9ec825172062aee51bc0ea57f4f323df113","unresolved":false,"context_lines":[{"line_number":41,"context_line":"addition, hugepages are assigned from memory and benefit from the same"},{"line_number":42,"context_line":"performance improvements as memory allocated using standard pages, thus they"},{"line_number":43,"context_line":"should be local also. Finally, PCI devices are directly associated with"},{"line_number":44,"context_line":"specific NUMA nodes for the purposes of DMA, so when using PCI device"},{"line_number":45,"context_line":"assignment it is also desirable that the guest be placed on the same NUMA node"},{"line_number":46,"context_line":"as any PCI device that is assigned to it."},{"line_number":47,"context_line":""}],"source_content_type":"text/x-rst","patch_set":1,"id":"1a122d0e_3fd1fefb","line":44,"updated":"2016-05-05 16:12:54.000000000","message":"The sentence might be too long.  consider breaking it after the coma: \"So when the PCI ...\"","commit_id":"78f32244badbaef989e826bf02fac2c14b7a6041"},{"author":{"_account_id":15334,"name":"Stephen Finucane","display_name":"stephenfin","email":"stephenfin@redhat.com","username":"sfinucan"},"change_message_id":"de849e581797cb26a10c9897f57d162af984b387","unresolved":false,"context_lines":[{"line_number":41,"context_line":"addition, hugepages are assigned from memory and benefit from the same"},{"line_number":42,"context_line":"performance improvements as memory allocated using standard pages, thus they"},{"line_number":43,"context_line":"should be local also. Finally, PCI devices are directly associated with"},{"line_number":44,"context_line":"specific NUMA nodes for the purposes of DMA, so when using PCI device"},{"line_number":45,"context_line":"assignment it is also desirable that the guest be placed on the same NUMA node"},{"line_number":46,"context_line":"as any PCI device that is assigned to it."},{"line_number":47,"context_line":""}],"source_content_type":"text/x-rst","patch_set":1,"id":"dab17558_797ce920","line":44,"in_reply_to":"1a122d0e_3fd1fefb","updated":"2016-05-06 10:05:10.000000000","message":"Done","commit_id":"78f32244badbaef989e826bf02fac2c14b7a6041"},{"author":{"_account_id":14962,"name":"ologvinova","email":"ologvinova@mirantis.com","username":"ologvinova"},"change_message_id":"1cdd9c80f23b6c6e5f9852adc8175e3929f73340","unresolved":false,"context_lines":[{"line_number":42,"context_line":"performance improvements as memory allocated using standard pages, thus they"},{"line_number":43,"context_line":"should be local also. Finally, PCI devices are directly associated with"},{"line_number":44,"context_line":"specific NUMA nodes for the purposes of DMA, so when using PCI device"},{"line_number":45,"context_line":"assignment it is also desirable that the guest be placed on the same NUMA node"},{"line_number":46,"context_line":"as any PCI device that is assigned to it."},{"line_number":47,"context_line":""},{"line_number":48,"context_line":"By default, if the RAM/vCPUs associated with a flavor can fit into a single"}],"source_content_type":"text/x-rst","patch_set":1,"id":"1a122d0e_53983b75","line":45,"range":{"start_line":45,"start_character":10,"end_line":45,"end_character":11},"updated":"2016-05-05 15:27:58.000000000","message":"comma","commit_id":"78f32244badbaef989e826bf02fac2c14b7a6041"},{"author":{"_account_id":15334,"name":"Stephen Finucane","display_name":"stephenfin","email":"stephenfin@redhat.com","username":"sfinucan"},"change_message_id":"de849e581797cb26a10c9897f57d162af984b387","unresolved":false,"context_lines":[{"line_number":42,"context_line":"performance improvements as memory allocated using standard pages, thus they"},{"line_number":43,"context_line":"should be local also. Finally, PCI devices are directly associated with"},{"line_number":44,"context_line":"specific NUMA nodes for the purposes of DMA, so when using PCI device"},{"line_number":45,"context_line":"assignment it is also desirable that the guest be placed on the same NUMA node"},{"line_number":46,"context_line":"as any PCI device that is assigned to it."},{"line_number":47,"context_line":""},{"line_number":48,"context_line":"By default, if the RAM/vCPUs associated with a flavor can fit into a single"}],"source_content_type":"text/x-rst","patch_set":1,"id":"dab17558_b976513c","line":45,"range":{"start_line":45,"start_character":10,"end_line":45,"end_character":11},"in_reply_to":"1a122d0e_53983b75","updated":"2016-05-06 10:05:10.000000000","message":"Done","commit_id":"78f32244badbaef989e826bf02fac2c14b7a6041"},{"author":{"_account_id":14962,"name":"ologvinova","email":"ologvinova@mirantis.com","username":"ologvinova"},"change_message_id":"1cdd9c80f23b6c6e5f9852adc8175e3929f73340","unresolved":false,"context_lines":[{"line_number":58,"context_line":""},{"line_number":59,"context_line":"   The NUMA topology shown in the instance will not necessarily reflect the"},{"line_number":60,"context_line":"   underlying topology of the instance. This is by design. See this bug"},{"line_number":61,"context_line":"   for further information. https://bugs.launchpad.net/nova/+bug/1466780"},{"line_number":62,"context_line":""},{"line_number":63,"context_line":"Custom NUMA Toplogies"},{"line_number":64,"context_line":"~~~~~~~~~~~~~~~~~~~~~"}],"source_content_type":"text/x-rst","patch_set":1,"id":"1a122d0e_b3ed4f90","line":61,"range":{"start_line":61,"start_character":26,"end_line":61,"end_character":27},"updated":"2016-05-05 15:27:58.000000000","message":"Please replace a period with a colon.","commit_id":"78f32244badbaef989e826bf02fac2c14b7a6041"},{"author":{"_account_id":15334,"name":"Stephen Finucane","display_name":"stephenfin","email":"stephenfin@redhat.com","username":"sfinucan"},"change_message_id":"de849e581797cb26a10c9897f57d162af984b387","unresolved":false,"context_lines":[{"line_number":58,"context_line":""},{"line_number":59,"context_line":"   The NUMA topology shown in the instance will not necessarily reflect the"},{"line_number":60,"context_line":"   underlying topology of the instance. This is by design. See this bug"},{"line_number":61,"context_line":"   for further information. https://bugs.launchpad.net/nova/+bug/1466780"},{"line_number":62,"context_line":""},{"line_number":63,"context_line":"Custom NUMA Toplogies"},{"line_number":64,"context_line":"~~~~~~~~~~~~~~~~~~~~~"}],"source_content_type":"text/x-rst","patch_set":1,"id":"dab17558_99ef0db0","line":61,"range":{"start_line":61,"start_character":26,"end_line":61,"end_character":27},"in_reply_to":"1a122d0e_84ba8b09","updated":"2016-05-06 10:05:10.000000000","message":"Done","commit_id":"78f32244badbaef989e826bf02fac2c14b7a6041"},{"author":{"_account_id":19779,"name":"Chason Chan","display_name":"chenxing","email":"chason.chan@foxmail.com","username":"chenxing"},"change_message_id":"cb2d23d337d283cf9c50987eb0bd6446ce424525","unresolved":false,"context_lines":[{"line_number":58,"context_line":""},{"line_number":59,"context_line":"   The NUMA topology shown in the instance will not necessarily reflect the"},{"line_number":60,"context_line":"   underlying topology of the instance. This is by design. See this bug"},{"line_number":61,"context_line":"   for further information. https://bugs.launchpad.net/nova/+bug/1466780"},{"line_number":62,"context_line":""},{"line_number":63,"context_line":"Custom NUMA Toplogies"},{"line_number":64,"context_line":"~~~~~~~~~~~~~~~~~~~~~"}],"source_content_type":"text/x-rst","patch_set":1,"id":"1a122d0e_e4d1bfcd","line":61,"range":{"start_line":61,"start_character":26,"end_line":61,"end_character":27},"in_reply_to":"1a122d0e_b3ed4f90","updated":"2016-05-05 15:56:20.000000000","message":"I have another suggestion:\nsee this `bug \u003chttps://bugs.launchpad.net/nova/+bug/1466780\u003e`__ for further information.","commit_id":"78f32244badbaef989e826bf02fac2c14b7a6041"},{"author":{"_account_id":17973,"name":"Ludovic Beliveau","email":"lbeliveau@gmail.com","username":"lbeliveau"},"change_message_id":"8968f9ec825172062aee51bc0ea57f4f323df113","unresolved":false,"context_lines":[{"line_number":58,"context_line":""},{"line_number":59,"context_line":"   The NUMA topology shown in the instance will not necessarily reflect the"},{"line_number":60,"context_line":"   underlying topology of the instance. This is by design. See this bug"},{"line_number":61,"context_line":"   for further information. https://bugs.launchpad.net/nova/+bug/1466780"},{"line_number":62,"context_line":""},{"line_number":63,"context_line":"Custom NUMA Toplogies"},{"line_number":64,"context_line":"~~~~~~~~~~~~~~~~~~~~~"}],"source_content_type":"text/x-rst","patch_set":1,"id":"1a122d0e_84ba8b09","line":61,"range":{"start_line":61,"start_character":26,"end_line":61,"end_character":27},"in_reply_to":"1a122d0e_b3ed4f90","updated":"2016-05-05 16:12:54.000000000","message":"Please bundle this web link with `\u003c\u003e`_.","commit_id":"78f32244badbaef989e826bf02fac2c14b7a6041"},{"author":{"_account_id":19779,"name":"Chason Chan","display_name":"chenxing","email":"chason.chan@foxmail.com","username":"chenxing"},"change_message_id":"cb2d23d337d283cf9c50987eb0bd6446ce424525","unresolved":false,"context_lines":[{"line_number":63,"context_line":"Custom NUMA Toplogies"},{"line_number":64,"context_line":"~~~~~~~~~~~~~~~~~~~~~"},{"line_number":65,"context_line":""},{"line_number":66,"context_line":"It is possible to override both the allocation of ..."},{"line_number":67,"context_line":""},{"line_number":68,"context_line":"Some workloads have very demanding requirements for memory access latency"},{"line_number":69,"context_line":"and/or bandwidth, which exceed that which is available from a single NUMA node"}],"source_content_type":"text/x-rst","patch_set":1,"id":"1a122d0e_a48ce735","line":66,"range":{"start_line":66,"start_character":49,"end_line":66,"end_character":53},"updated":"2016-05-05 15:56:20.000000000","message":"what is the meaning?","commit_id":"78f32244badbaef989e826bf02fac2c14b7a6041"},{"author":{"_account_id":15334,"name":"Stephen Finucane","display_name":"stephenfin","email":"stephenfin@redhat.com","username":"sfinucan"},"change_message_id":"de849e581797cb26a10c9897f57d162af984b387","unresolved":false,"context_lines":[{"line_number":63,"context_line":"Custom NUMA Toplogies"},{"line_number":64,"context_line":"~~~~~~~~~~~~~~~~~~~~~"},{"line_number":65,"context_line":""},{"line_number":66,"context_line":"It is possible to override both the allocation of ..."},{"line_number":67,"context_line":""},{"line_number":68,"context_line":"Some workloads have very demanding requirements for memory access latency"},{"line_number":69,"context_line":"and/or bandwidth, which exceed that which is available from a single NUMA node"}],"source_content_type":"text/x-rst","patch_set":1,"id":"dab17558_b9f2d198","line":66,"range":{"start_line":66,"start_character":49,"end_line":66,"end_character":53},"in_reply_to":"1a122d0e_a48ce735","updated":"2016-05-06 10:05:10.000000000","message":"This is still WIP. I haven\u0027t finished this section","commit_id":"78f32244badbaef989e826bf02fac2c14b7a6041"},{"author":{"_account_id":14962,"name":"ologvinova","email":"ologvinova@mirantis.com","username":"ologvinova"},"change_message_id":"1cdd9c80f23b6c6e5f9852adc8175e3929f73340","unresolved":false,"context_lines":[{"line_number":66,"context_line":"It is possible to override both the allocation of ..."},{"line_number":67,"context_line":""},{"line_number":68,"context_line":"Some workloads have very demanding requirements for memory access latency"},{"line_number":69,"context_line":"and/or bandwidth, which exceed that which is available from a single NUMA node"},{"line_number":70,"context_line":"For such workloads, it is be beneficial to spread the instance across multiple"},{"line_number":71,"context_line":"host NUMA nodes, even if the instance\u0027s RAM/vCPUs could theoretically fit on a"},{"line_number":72,"context_line":"single NUMA node."}],"source_content_type":"text/x-rst","patch_set":1,"id":"1a122d0e_732bb795","line":69,"range":{"start_line":69,"start_character":77,"end_line":69,"end_character":78},"updated":"2016-05-05 15:27:58.000000000","message":"A full stop at the end of sentence, please.","commit_id":"78f32244badbaef989e826bf02fac2c14b7a6041"},{"author":{"_account_id":14962,"name":"ologvinova","email":"ologvinova@mirantis.com","username":"ologvinova"},"change_message_id":"1cdd9c80f23b6c6e5f9852adc8175e3929f73340","unresolved":false,"context_lines":[{"line_number":66,"context_line":"It is possible to override both the allocation of ..."},{"line_number":67,"context_line":""},{"line_number":68,"context_line":"Some workloads have very demanding requirements for memory access latency"},{"line_number":69,"context_line":"and/or bandwidth, which exceed that which is available from a single NUMA node"},{"line_number":70,"context_line":"For such workloads, it is be beneficial to spread the instance across multiple"},{"line_number":71,"context_line":"host NUMA nodes, even if the instance\u0027s RAM/vCPUs could theoretically fit on a"},{"line_number":72,"context_line":"single NUMA node."}],"source_content_type":"text/x-rst","patch_set":1,"id":"1a122d0e_9335b3f8","line":69,"range":{"start_line":69,"start_character":18,"end_line":69,"end_character":41},"updated":"2016-05-05 15:27:58.000000000","message":"Could you please rephrase this part?","commit_id":"78f32244badbaef989e826bf02fac2c14b7a6041"},{"author":{"_account_id":15334,"name":"Stephen Finucane","display_name":"stephenfin","email":"stephenfin@redhat.com","username":"sfinucan"},"change_message_id":"de849e581797cb26a10c9897f57d162af984b387","unresolved":false,"context_lines":[{"line_number":66,"context_line":"It is possible to override both the allocation of ..."},{"line_number":67,"context_line":""},{"line_number":68,"context_line":"Some workloads have very demanding requirements for memory access latency"},{"line_number":69,"context_line":"and/or bandwidth, which exceed that which is available from a single NUMA node"},{"line_number":70,"context_line":"For such workloads, it is be beneficial to spread the instance across multiple"},{"line_number":71,"context_line":"host NUMA nodes, even if the instance\u0027s RAM/vCPUs could theoretically fit on a"},{"line_number":72,"context_line":"single NUMA node."}],"source_content_type":"text/x-rst","patch_set":1,"id":"dab17558_f95f3973","line":69,"range":{"start_line":69,"start_character":77,"end_line":69,"end_character":78},"in_reply_to":"1a122d0e_732bb795","updated":"2016-05-06 10:05:10.000000000","message":"Done","commit_id":"78f32244badbaef989e826bf02fac2c14b7a6041"},{"author":{"_account_id":15334,"name":"Stephen Finucane","display_name":"stephenfin","email":"stephenfin@redhat.com","username":"sfinucan"},"change_message_id":"de849e581797cb26a10c9897f57d162af984b387","unresolved":false,"context_lines":[{"line_number":66,"context_line":"It is possible to override both the allocation of ..."},{"line_number":67,"context_line":""},{"line_number":68,"context_line":"Some workloads have very demanding requirements for memory access latency"},{"line_number":69,"context_line":"and/or bandwidth, which exceed that which is available from a single NUMA node"},{"line_number":70,"context_line":"For such workloads, it is be beneficial to spread the instance across multiple"},{"line_number":71,"context_line":"host NUMA nodes, even if the instance\u0027s RAM/vCPUs could theoretically fit on a"},{"line_number":72,"context_line":"single NUMA node."}],"source_content_type":"text/x-rst","patch_set":1,"id":"dab17558_d964f54a","line":69,"range":{"start_line":69,"start_character":18,"end_line":69,"end_character":41},"in_reply_to":"1a122d0e_9335b3f8","updated":"2016-05-06 10:05:10.000000000","message":"Done","commit_id":"78f32244badbaef989e826bf02fac2c14b7a6041"},{"author":{"_account_id":14962,"name":"ologvinova","email":"ologvinova@mirantis.com","username":"ologvinova"},"change_message_id":"1cdd9c80f23b6c6e5f9852adc8175e3929f73340","unresolved":false,"context_lines":[{"line_number":67,"context_line":""},{"line_number":68,"context_line":"Some workloads have very demanding requirements for memory access latency"},{"line_number":69,"context_line":"and/or bandwidth, which exceed that which is available from a single NUMA node"},{"line_number":70,"context_line":"For such workloads, it is be beneficial to spread the instance across multiple"},{"line_number":71,"context_line":"host NUMA nodes, even if the instance\u0027s RAM/vCPUs could theoretically fit on a"},{"line_number":72,"context_line":"single NUMA node."},{"line_number":73,"context_line":""}],"source_content_type":"text/x-rst","patch_set":1,"id":"1a122d0e_d3b60b88","line":70,"range":{"start_line":70,"start_character":26,"end_line":70,"end_character":28},"updated":"2016-05-05 15:27:58.000000000","message":"Remove \u0027be\u0027.","commit_id":"78f32244badbaef989e826bf02fac2c14b7a6041"},{"author":{"_account_id":15334,"name":"Stephen Finucane","display_name":"stephenfin","email":"stephenfin@redhat.com","username":"sfinucan"},"change_message_id":"de849e581797cb26a10c9897f57d162af984b387","unresolved":false,"context_lines":[{"line_number":67,"context_line":""},{"line_number":68,"context_line":"Some workloads have very demanding requirements for memory access latency"},{"line_number":69,"context_line":"and/or bandwidth, which exceed that which is available from a single NUMA node"},{"line_number":70,"context_line":"For such workloads, it is be beneficial to spread the instance across multiple"},{"line_number":71,"context_line":"host NUMA nodes, even if the instance\u0027s RAM/vCPUs could theoretically fit on a"},{"line_number":72,"context_line":"single NUMA node."},{"line_number":73,"context_line":""}],"source_content_type":"text/x-rst","patch_set":1,"id":"dab17558_b9777101","line":70,"range":{"start_line":70,"start_character":26,"end_line":70,"end_character":28},"in_reply_to":"1a122d0e_d3b60b88","updated":"2016-05-06 10:05:10.000000000","message":"Done","commit_id":"78f32244badbaef989e826bf02fac2c14b7a6041"},{"author":{"_account_id":14962,"name":"ologvinova","email":"ologvinova@mirantis.com","username":"ologvinova"},"change_message_id":"1cdd9c80f23b6c6e5f9852adc8175e3929f73340","unresolved":false,"context_lines":[{"line_number":77,"context_line":"~~~~~~~~~~~~"},{"line_number":78,"context_line":""},{"line_number":79,"context_line":"By default, the Compute service permits overcommiting of instance CPUs. This"},{"line_number":80,"context_line":"is achived by allowing the instance CPU processes to float across host CPUs."},{"line_number":81,"context_line":"In heavily contended system, this provides optimal system performance at the"},{"line_number":82,"context_line":"expense of performance and latency for individual instances. However, some"},{"line_number":83,"context_line":"applications must exhibit real-time or near real-time behavior; notably those"}],"source_content_type":"text/x-rst","patch_set":1,"id":"1a122d0e_c6092b2d","line":80,"range":{"start_line":80,"start_character":3,"end_line":80,"end_character":10},"updated":"2016-05-05 15:27:58.000000000","message":"achieved","commit_id":"78f32244badbaef989e826bf02fac2c14b7a6041"},{"author":{"_account_id":15334,"name":"Stephen Finucane","display_name":"stephenfin","email":"stephenfin@redhat.com","username":"sfinucan"},"change_message_id":"de849e581797cb26a10c9897f57d162af984b387","unresolved":false,"context_lines":[{"line_number":77,"context_line":"~~~~~~~~~~~~"},{"line_number":78,"context_line":""},{"line_number":79,"context_line":"By default, the Compute service permits overcommiting of instance CPUs. This"},{"line_number":80,"context_line":"is achived by allowing the instance CPU processes to float across host CPUs."},{"line_number":81,"context_line":"In heavily contended system, this provides optimal system performance at the"},{"line_number":82,"context_line":"expense of performance and latency for individual instances. However, some"},{"line_number":83,"context_line":"applications must exhibit real-time or near real-time behavior; notably those"}],"source_content_type":"text/x-rst","patch_set":1,"id":"dab17558_994b8db0","line":80,"range":{"start_line":80,"start_character":3,"end_line":80,"end_character":10},"in_reply_to":"1a122d0e_c6092b2d","updated":"2016-05-06 10:05:10.000000000","message":"Done","commit_id":"78f32244badbaef989e826bf02fac2c14b7a6041"},{"author":{"_account_id":19779,"name":"Chason Chan","display_name":"chenxing","email":"chason.chan@foxmail.com","username":"chenxing"},"change_message_id":"cb2d23d337d283cf9c50987eb0bd6446ce424525","unresolved":false,"context_lines":[{"line_number":80,"context_line":"is achived by allowing the instance CPU processes to float across host CPUs."},{"line_number":81,"context_line":"In heavily contended system, this provides optimal system performance at the"},{"line_number":82,"context_line":"expense of performance and latency for individual instances. However, some"},{"line_number":83,"context_line":"applications must exhibit real-time or near real-time behavior; notably those"},{"line_number":84,"context_line":"found in an NFV deployment."}],"source_content_type":"text/x-rst","patch_set":1,"id":"1a122d0e_84ee0bb3","line":83,"range":{"start_line":83,"start_character":62,"end_line":83,"end_character":64},"updated":"2016-05-05 15:56:20.000000000","message":"why do you use a colon here?","commit_id":"78f32244badbaef989e826bf02fac2c14b7a6041"},{"author":{"_account_id":15334,"name":"Stephen Finucane","display_name":"stephenfin","email":"stephenfin@redhat.com","username":"sfinucan"},"change_message_id":"de849e581797cb26a10c9897f57d162af984b387","unresolved":false,"context_lines":[{"line_number":80,"context_line":"is achived by allowing the instance CPU processes to float across host CPUs."},{"line_number":81,"context_line":"In heavily contended system, this provides optimal system performance at the"},{"line_number":82,"context_line":"expense of performance and latency for individual instances. However, some"},{"line_number":83,"context_line":"applications must exhibit real-time or near real-time behavior; notably those"},{"line_number":84,"context_line":"found in an NFV deployment."}],"source_content_type":"text/x-rst","patch_set":1,"id":"dab17558_b94e51bf","line":83,"range":{"start_line":83,"start_character":62,"end_line":83,"end_character":64},"in_reply_to":"1a122d0e_84ee0bb3","updated":"2016-05-06 10:05:10.000000000","message":"Done","commit_id":"78f32244badbaef989e826bf02fac2c14b7a6041"},{"author":{"_account_id":10897,"name":"Joseph Robinson","email":"josephremail@gmail.com","username":"RobiJoe"},"change_message_id":"7cb9be283b89eac49a14d4914faa91d79305fea6","unresolved":false,"context_lines":[{"line_number":3,"context_line":"\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d"},{"line_number":4,"context_line":""},{"line_number":5,"context_line":"OpenStack clouds run on platforms that differ greatly in the capabilities that"},{"line_number":6,"context_line":"they provide. By default, the Compute service seeks to abstract the underlying"},{"line_number":7,"context_line":"hardware that it runs on, rather than exposing specifics about the underlying"},{"line_number":8,"context_line":"host platforms. This abstraction manifests itself in many ways. For example,"},{"line_number":9,"context_line":"rather than exposing the types and topologies of CPUs running on hosts, the"}],"source_content_type":"text/x-rst","patch_set":2,"id":"dab17558_40960a4d","line":6,"updated":"2016-05-09 07:17:48.000000000","message":"Lower case \u0027c\u0027 for compute here - http://docs.openstack.org/contributor-guide/writing-style/openstack-components.html","commit_id":"a04f134929b090af1e90136b15d40200265d2543"},{"author":{"_account_id":15334,"name":"Stephen Finucane","display_name":"stephenfin","email":"stephenfin@redhat.com","username":"sfinucan"},"change_message_id":"14480b9ff6f9fa3abfc1bc35145a0cf1f287d401","unresolved":false,"context_lines":[{"line_number":3,"context_line":"\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d"},{"line_number":4,"context_line":""},{"line_number":5,"context_line":"OpenStack clouds run on platforms that differ greatly in the capabilities that"},{"line_number":6,"context_line":"they provide. By default, the Compute service seeks to abstract the underlying"},{"line_number":7,"context_line":"hardware that it runs on, rather than exposing specifics about the underlying"},{"line_number":8,"context_line":"host platforms. This abstraction manifests itself in many ways. For example,"},{"line_number":9,"context_line":"rather than exposing the types and topologies of CPUs running on hosts, the"}],"source_content_type":"text/x-rst","patch_set":2,"id":"dab17558_d64d9f73","line":6,"in_reply_to":"dab17558_40960a4d","updated":"2016-05-09 10:54:34.000000000","message":"Done","commit_id":"a04f134929b090af1e90136b15d40200265d2543"},{"author":{"_account_id":10897,"name":"Joseph Robinson","email":"josephremail@gmail.com","username":"RobiJoe"},"change_message_id":"7cb9be283b89eac49a14d4914faa91d79305fea6","unresolved":false,"context_lines":[{"line_number":11,"context_line":"for over-provisioning of these. In a similar manner, rather than exposing the"},{"line_number":12,"context_line":"individual types of network devices available on hosts, generic"},{"line_number":13,"context_line":"software-powered network ports are provided. These features are designed to"},{"line_number":14,"context_line":"enable high resource utilization and allows the service to provide a generic"},{"line_number":15,"context_line":"cost-effective and highly-scalable cloud upon which to build applications."},{"line_number":16,"context_line":""},{"line_number":17,"context_line":"This abstraction is beneficial for most applications. However, there are some"}],"source_content_type":"text/x-rst","patch_set":2,"id":"dab17558_80775239","line":14,"updated":"2016-05-09 07:17:48.000000000","message":"Switch to \u0027allow\u0027 here for verb subject agreement.","commit_id":"a04f134929b090af1e90136b15d40200265d2543"},{"author":{"_account_id":15334,"name":"Stephen Finucane","display_name":"stephenfin","email":"stephenfin@redhat.com","username":"sfinucan"},"change_message_id":"14480b9ff6f9fa3abfc1bc35145a0cf1f287d401","unresolved":false,"context_lines":[{"line_number":11,"context_line":"for over-provisioning of these. In a similar manner, rather than exposing the"},{"line_number":12,"context_line":"individual types of network devices available on hosts, generic"},{"line_number":13,"context_line":"software-powered network ports are provided. These features are designed to"},{"line_number":14,"context_line":"enable high resource utilization and allows the service to provide a generic"},{"line_number":15,"context_line":"cost-effective and highly-scalable cloud upon which to build applications."},{"line_number":16,"context_line":""},{"line_number":17,"context_line":"This abstraction is beneficial for most applications. However, there are some"}],"source_content_type":"text/x-rst","patch_set":2,"id":"dab17558_363d0bbc","line":14,"in_reply_to":"dab17558_80775239","updated":"2016-05-09 10:54:34.000000000","message":"Done","commit_id":"a04f134929b090af1e90136b15d40200265d2543"},{"author":{"_account_id":10897,"name":"Joseph Robinson","email":"josephremail@gmail.com","username":"RobiJoe"},"change_message_id":"7cb9be283b89eac49a14d4914faa91d79305fea6","unresolved":false,"context_lines":[{"line_number":17,"context_line":"This abstraction is beneficial for most applications. However, there are some"},{"line_number":18,"context_line":"applications where determinism and per-instance performance are important, if"},{"line_number":19,"context_line":"not vital. In these cases, instances can be expected to deliver near-native"},{"line_number":20,"context_line":"performance. The Compute service provides features to improve individual"},{"line_number":21,"context_line":"instance for these kind of applications."},{"line_number":22,"context_line":""},{"line_number":23,"context_line":"Guest NUMA topology"}],"source_content_type":"text/x-rst","patch_set":2,"id":"dab17558_c08f9a21","line":20,"updated":"2016-05-09 07:17:48.000000000","message":"lower case compute.","commit_id":"a04f134929b090af1e90136b15d40200265d2543"},{"author":{"_account_id":15334,"name":"Stephen Finucane","display_name":"stephenfin","email":"stephenfin@redhat.com","username":"sfinucan"},"change_message_id":"14480b9ff6f9fa3abfc1bc35145a0cf1f287d401","unresolved":false,"context_lines":[{"line_number":17,"context_line":"This abstraction is beneficial for most applications. However, there are some"},{"line_number":18,"context_line":"applications where determinism and per-instance performance are important, if"},{"line_number":19,"context_line":"not vital. In these cases, instances can be expected to deliver near-native"},{"line_number":20,"context_line":"performance. The Compute service provides features to improve individual"},{"line_number":21,"context_line":"instance for these kind of applications."},{"line_number":22,"context_line":""},{"line_number":23,"context_line":"Guest NUMA topology"}],"source_content_type":"text/x-rst","patch_set":2,"id":"dab17558_5669afbd","line":20,"in_reply_to":"dab17558_c08f9a21","updated":"2016-05-09 10:54:34.000000000","message":"Done","commit_id":"a04f134929b090af1e90136b15d40200265d2543"},{"author":{"_account_id":10897,"name":"Joseph Robinson","email":"josephremail@gmail.com","username":"RobiJoe"},"change_message_id":"7cb9be283b89eac49a14d4914faa91d79305fea6","unresolved":false,"context_lines":[{"line_number":47,"context_line":""},{"line_number":48,"context_line":"By default, if the RAM/vCPUs associated with a flavor can fit into a single"},{"line_number":49,"context_line":"NUMA node, then the vCPUs of that instance will only float across the host"},{"line_number":50,"context_line":"CPUs on a given node and the instance\u0027s NUMA topology will reflect a single"},{"line_number":51,"context_line":"node layout. If the RAM/vCPUs associated with a flavor are larger than any"},{"line_number":52,"context_line":"single NUMA node, then the instance vCPUs will float across a number of cores"},{"line_number":53,"context_line":"and the instance\u0027s NUMA topology will reflect a multi-node layout. This allows"}],"source_content_type":"text/x-rst","patch_set":2,"id":"dab17558_e0b3bee7","line":50,"updated":"2016-05-09 07:17:48.000000000","message":"change to \u0027instance NUMA topology\u0027 - removing the apostrophe.","commit_id":"a04f134929b090af1e90136b15d40200265d2543"},{"author":{"_account_id":15334,"name":"Stephen Finucane","display_name":"stephenfin","email":"stephenfin@redhat.com","username":"sfinucan"},"change_message_id":"14480b9ff6f9fa3abfc1bc35145a0cf1f287d401","unresolved":false,"context_lines":[{"line_number":47,"context_line":""},{"line_number":48,"context_line":"By default, if the RAM/vCPUs associated with a flavor can fit into a single"},{"line_number":49,"context_line":"NUMA node, then the vCPUs of that instance will only float across the host"},{"line_number":50,"context_line":"CPUs on a given node and the instance\u0027s NUMA topology will reflect a single"},{"line_number":51,"context_line":"node layout. If the RAM/vCPUs associated with a flavor are larger than any"},{"line_number":52,"context_line":"single NUMA node, then the instance vCPUs will float across a number of cores"},{"line_number":53,"context_line":"and the instance\u0027s NUMA topology will reflect a multi-node layout. This allows"}],"source_content_type":"text/x-rst","patch_set":2,"id":"dab17558_3664ebe3","line":50,"in_reply_to":"dab17558_e0b3bee7","updated":"2016-05-09 10:54:34.000000000","message":"Done","commit_id":"a04f134929b090af1e90136b15d40200265d2543"},{"author":{"_account_id":10897,"name":"Joseph Robinson","email":"josephremail@gmail.com","username":"RobiJoe"},"change_message_id":"7cb9be283b89eac49a14d4914faa91d79305fea6","unresolved":false,"context_lines":[{"line_number":48,"context_line":"By default, if the RAM/vCPUs associated with a flavor can fit into a single"},{"line_number":49,"context_line":"NUMA node, then the vCPUs of that instance will only float across the host"},{"line_number":50,"context_line":"CPUs on a given node and the instance\u0027s NUMA topology will reflect a single"},{"line_number":51,"context_line":"node layout. If the RAM/vCPUs associated with a flavor are larger than any"},{"line_number":52,"context_line":"single NUMA node, then the instance vCPUs will float across a number of cores"},{"line_number":53,"context_line":"and the instance\u0027s NUMA topology will reflect a multi-node layout. This allows"},{"line_number":54,"context_line":"the operating systems of the instance to intelligently schedule the workloads"}],"source_content_type":"text/x-rst","patch_set":2,"id":"dab17558_40d28a01","line":51,"updated":"2016-05-09 07:17:48.000000000","message":"switch out the / here for \u0027and\u0027, \u0027or\u0027.","commit_id":"a04f134929b090af1e90136b15d40200265d2543"},{"author":{"_account_id":15334,"name":"Stephen Finucane","display_name":"stephenfin","email":"stephenfin@redhat.com","username":"sfinucan"},"change_message_id":"14480b9ff6f9fa3abfc1bc35145a0cf1f287d401","unresolved":false,"context_lines":[{"line_number":48,"context_line":"By default, if the RAM/vCPUs associated with a flavor can fit into a single"},{"line_number":49,"context_line":"NUMA node, then the vCPUs of that instance will only float across the host"},{"line_number":50,"context_line":"CPUs on a given node and the instance\u0027s NUMA topology will reflect a single"},{"line_number":51,"context_line":"node layout. If the RAM/vCPUs associated with a flavor are larger than any"},{"line_number":52,"context_line":"single NUMA node, then the instance vCPUs will float across a number of cores"},{"line_number":53,"context_line":"and the instance\u0027s NUMA topology will reflect a multi-node layout. This allows"},{"line_number":54,"context_line":"the operating systems of the instance to intelligently schedule the workloads"}],"source_content_type":"text/x-rst","patch_set":2,"id":"dab17558_966317dc","line":51,"in_reply_to":"dab17558_40d28a01","updated":"2016-05-09 10:54:34.000000000","message":"Done","commit_id":"a04f134929b090af1e90136b15d40200265d2543"},{"author":{"_account_id":10897,"name":"Joseph Robinson","email":"josephremail@gmail.com","username":"RobiJoe"},"change_message_id":"7cb9be283b89eac49a14d4914faa91d79305fea6","unresolved":false,"context_lines":[{"line_number":50,"context_line":"CPUs on a given node and the instance\u0027s NUMA topology will reflect a single"},{"line_number":51,"context_line":"node layout. If the RAM/vCPUs associated with a flavor are larger than any"},{"line_number":52,"context_line":"single NUMA node, then the instance vCPUs will float across a number of cores"},{"line_number":53,"context_line":"and the instance\u0027s NUMA topology will reflect a multi-node layout. This allows"},{"line_number":54,"context_line":"the operating systems of the instance to intelligently schedule the workloads"},{"line_number":55,"context_line":"that it runs."},{"line_number":56,"context_line":""}],"source_content_type":"text/x-rst","patch_set":2,"id":"dab17558_20cdc65d","line":53,"updated":"2016-05-09 07:17:48.000000000","message":"instance NUMA topology.","commit_id":"a04f134929b090af1e90136b15d40200265d2543"},{"author":{"_account_id":15334,"name":"Stephen Finucane","display_name":"stephenfin","email":"stephenfin@redhat.com","username":"sfinucan"},"change_message_id":"14480b9ff6f9fa3abfc1bc35145a0cf1f287d401","unresolved":false,"context_lines":[{"line_number":50,"context_line":"CPUs on a given node and the instance\u0027s NUMA topology will reflect a single"},{"line_number":51,"context_line":"node layout. If the RAM/vCPUs associated with a flavor are larger than any"},{"line_number":52,"context_line":"single NUMA node, then the instance vCPUs will float across a number of cores"},{"line_number":53,"context_line":"and the instance\u0027s NUMA topology will reflect a multi-node layout. This allows"},{"line_number":54,"context_line":"the operating systems of the instance to intelligently schedule the workloads"},{"line_number":55,"context_line":"that it runs."},{"line_number":56,"context_line":""}],"source_content_type":"text/x-rst","patch_set":2,"id":"dab17558_76b5d32f","line":53,"in_reply_to":"dab17558_20cdc65d","updated":"2016-05-09 10:54:34.000000000","message":"Done","commit_id":"a04f134929b090af1e90136b15d40200265d2543"},{"author":{"_account_id":10897,"name":"Joseph Robinson","email":"josephremail@gmail.com","username":"RobiJoe"},"change_message_id":"7cb9be283b89eac49a14d4914faa91d79305fea6","unresolved":false,"context_lines":[{"line_number":60,"context_line":"   underlying topology of the instance. This is by design. See this"},{"line_number":61,"context_line":"   `bug \u003chttps://bugs.launchpad.net/nova/+bug/1466780\u003e`_ for more information."},{"line_number":62,"context_line":""},{"line_number":63,"context_line":"Custom NUMA topologies"},{"line_number":64,"context_line":"~~~~~~~~~~~~~~~~~~~~~"},{"line_number":65,"context_line":""},{"line_number":66,"context_line":"It is possible to override both the allocation of ..."}],"source_content_type":"text/x-rst","patch_set":2,"id":"dab17558_00ffe267","line":63,"updated":"2016-05-09 07:17:48.000000000","message":"Comment out this section if it is not ready.","commit_id":"a04f134929b090af1e90136b15d40200265d2543"},{"author":{"_account_id":15334,"name":"Stephen Finucane","display_name":"stephenfin","email":"stephenfin@redhat.com","username":"sfinucan"},"change_message_id":"14480b9ff6f9fa3abfc1bc35145a0cf1f287d401","unresolved":false,"context_lines":[{"line_number":60,"context_line":"   underlying topology of the instance. This is by design. See this"},{"line_number":61,"context_line":"   `bug \u003chttps://bugs.launchpad.net/nova/+bug/1466780\u003e`_ for more information."},{"line_number":62,"context_line":""},{"line_number":63,"context_line":"Custom NUMA topologies"},{"line_number":64,"context_line":"~~~~~~~~~~~~~~~~~~~~~"},{"line_number":65,"context_line":""},{"line_number":66,"context_line":"It is possible to override both the allocation of ..."}],"source_content_type":"text/x-rst","patch_set":2,"id":"dab17558_b6857b5a","line":63,"in_reply_to":"dab17558_00ffe267","updated":"2016-05-09 10:54:34.000000000","message":"Completed in a follow up patch","commit_id":"a04f134929b090af1e90136b15d40200265d2543"},{"author":{"_account_id":10897,"name":"Joseph Robinson","email":"josephremail@gmail.com","username":"RobiJoe"},"change_message_id":"7cb9be283b89eac49a14d4914faa91d79305fea6","unresolved":false,"context_lines":[{"line_number":76,"context_line":"CPU policies"},{"line_number":77,"context_line":"~~~~~~~~~~~~"},{"line_number":78,"context_line":""},{"line_number":79,"context_line":"By default, the Compute service permits overcommiting of instance CPUs. This"},{"line_number":80,"context_line":"is achieved by allowing the instance CPU processes to float across host CPUs."},{"line_number":81,"context_line":"In heavily contended system, this provides optimal system performance at the"},{"line_number":82,"context_line":"expense of performance and latency for individual instances. However, some"}],"source_content_type":"text/x-rst","patch_set":2,"id":"dab17558_20f4a685","line":79,"updated":"2016-05-09 07:17:48.000000000","message":"compute service.","commit_id":"a04f134929b090af1e90136b15d40200265d2543"},{"author":{"_account_id":15334,"name":"Stephen Finucane","display_name":"stephenfin","email":"stephenfin@redhat.com","username":"sfinucan"},"change_message_id":"14480b9ff6f9fa3abfc1bc35145a0cf1f287d401","unresolved":false,"context_lines":[{"line_number":76,"context_line":"CPU policies"},{"line_number":77,"context_line":"~~~~~~~~~~~~"},{"line_number":78,"context_line":""},{"line_number":79,"context_line":"By default, the Compute service permits overcommiting of instance CPUs. This"},{"line_number":80,"context_line":"is achieved by allowing the instance CPU processes to float across host CPUs."},{"line_number":81,"context_line":"In heavily contended system, this provides optimal system performance at the"},{"line_number":82,"context_line":"expense of performance and latency for individual instances. However, some"}],"source_content_type":"text/x-rst","patch_set":2,"id":"dab17558_768cb37f","line":79,"in_reply_to":"dab17558_20f4a685","updated":"2016-05-09 10:54:34.000000000","message":"Done","commit_id":"a04f134929b090af1e90136b15d40200265d2543"},{"author":{"_account_id":10897,"name":"Joseph Robinson","email":"josephremail@gmail.com","username":"RobiJoe"},"change_message_id":"7cb9be283b89eac49a14d4914faa91d79305fea6","unresolved":false,"context_lines":[{"line_number":78,"context_line":""},{"line_number":79,"context_line":"By default, the Compute service permits overcommiting of instance CPUs. This"},{"line_number":80,"context_line":"is achieved by allowing the instance CPU processes to float across host CPUs."},{"line_number":81,"context_line":"In heavily contended system, this provides optimal system performance at the"},{"line_number":82,"context_line":"expense of performance and latency for individual instances. However, some"},{"line_number":83,"context_line":"applications must exhibit real-time or near real-time behavior notably those"},{"line_number":84,"context_line":"found in an NFV deployment."}],"source_content_type":"text/x-rst","patch_set":2,"id":"dab17558_804ab2c3","line":81,"updated":"2016-05-09 07:17:48.000000000","message":"switch to plural - \u0027In heavily contended systems, ..\u0027","commit_id":"a04f134929b090af1e90136b15d40200265d2543"},{"author":{"_account_id":15334,"name":"Stephen Finucane","display_name":"stephenfin","email":"stephenfin@redhat.com","username":"sfinucan"},"change_message_id":"14480b9ff6f9fa3abfc1bc35145a0cf1f287d401","unresolved":false,"context_lines":[{"line_number":78,"context_line":""},{"line_number":79,"context_line":"By default, the Compute service permits overcommiting of instance CPUs. This"},{"line_number":80,"context_line":"is achieved by allowing the instance CPU processes to float across host CPUs."},{"line_number":81,"context_line":"In heavily contended system, this provides optimal system performance at the"},{"line_number":82,"context_line":"expense of performance and latency for individual instances. However, some"},{"line_number":83,"context_line":"applications must exhibit real-time or near real-time behavior notably those"},{"line_number":84,"context_line":"found in an NFV deployment."}],"source_content_type":"text/x-rst","patch_set":2,"id":"dab17558_9691579c","line":81,"in_reply_to":"dab17558_804ab2c3","updated":"2016-05-09 10:54:34.000000000","message":"Done","commit_id":"a04f134929b090af1e90136b15d40200265d2543"},{"author":{"_account_id":17973,"name":"Ludovic Beliveau","email":"lbeliveau@gmail.com","username":"lbeliveau"},"change_message_id":"eaad34e066c8beb6845264d43d990ed02d882246","unresolved":false,"context_lines":[{"line_number":28,"context_line":"   The functionality described below is currently only supported by the"},{"line_number":29,"context_line":"   libvirt driver."},{"line_number":30,"context_line":""},{"line_number":31,"context_line":"Non-uniform memory access (NUMA) is a design found in many modern,"},{"line_number":32,"context_line":"multi-socket systems. In a NUMA system, system memory is divided into cells"},{"line_number":33,"context_line":"or nodes that are associated with particular CPUs. Requests for memory on"},{"line_number":34,"context_line":"other nodes are possible through an interconnect bus. However, bandwidth"}],"source_content_type":"text/x-rst","patch_set":3,"id":"dab17558_9226bf14","line":31,"updated":"2016-05-06 13:24:17.000000000","message":"There will be quite a lot of acronyms used throughout the text.  I\u0027m wondering if putting them in a table makes sense.\n\nSee http://docs.openstack.org/mitaka/networking-guide/adv-config-sriov.html.\n\nReally up to you !","commit_id":"d9cad91e3c20e79da095040dc1eeff180b2d172b"},{"author":{"_account_id":15334,"name":"Stephen Finucane","display_name":"stephenfin","email":"stephenfin@redhat.com","username":"sfinucan"},"change_message_id":"216479b3bf6b41ef23aec1a14b1272cad274268d","unresolved":false,"context_lines":[{"line_number":28,"context_line":"   The functionality described below is currently only supported by the"},{"line_number":29,"context_line":"   libvirt driver."},{"line_number":30,"context_line":""},{"line_number":31,"context_line":"Non-uniform memory access (NUMA) is a design found in many modern,"},{"line_number":32,"context_line":"multi-socket systems. In a NUMA system, system memory is divided into cells"},{"line_number":33,"context_line":"or nodes that are associated with particular CPUs. Requests for memory on"},{"line_number":34,"context_line":"other nodes are possible through an interconnect bus. However, bandwidth"}],"source_content_type":"text/x-rst","patch_set":3,"id":"dab17558_019d5665","line":31,"in_reply_to":"dab17558_9226bf14","updated":"2016-05-06 15:18:44.000000000","message":"Good idea. I think I\u0027ll do that.","commit_id":"d9cad91e3c20e79da095040dc1eeff180b2d172b"},{"author":{"_account_id":17973,"name":"Ludovic Beliveau","email":"lbeliveau@gmail.com","username":"lbeliveau"},"change_message_id":"eaad34e066c8beb6845264d43d990ed02d882246","unresolved":false,"context_lines":[{"line_number":86,"context_line":"       --property hw:numa_cpus.0\u003d0,1 \\"},{"line_number":87,"context_line":"       --property hw:numa_cpus.1\u003d2,3,4,5,6,7 \\"},{"line_number":88,"context_line":"       --property hw:numa_mem.0\u003d2 \\"},{"line_number":89,"context_line":"       --property hw:numa_mem.1\u003d6"},{"line_number":90,"context_line":""},{"line_number":91,"context_line":"The NUMA topology exposed inside the guest can be overriden. To configure a"},{"line_number":92,"context_line":"flavor with eight cores to provide a two socket, two core and two thread"}],"source_content_type":"text/x-rst","patch_set":3,"id":"dab17558_edbacabf","line":89,"updated":"2016-05-06 13:24:17.000000000","message":"It\u0027s not clear what this example is about.  IMHO it needs additional text to explain each properties and what will be the effect on the guest.\n\nLooks like http://docs.openstack.org/admin-guide/compute-flavors.html is not complete wrt numa.  Maybe that can be explained in the flavor ...\n\nMaybe this should be addressed.  Then a link should be provided here.","commit_id":"d9cad91e3c20e79da095040dc1eeff180b2d172b"},{"author":{"_account_id":15334,"name":"Stephen Finucane","display_name":"stephenfin","email":"stephenfin@redhat.com","username":"sfinucan"},"change_message_id":"216479b3bf6b41ef23aec1a14b1272cad274268d","unresolved":false,"context_lines":[{"line_number":86,"context_line":"       --property hw:numa_cpus.0\u003d0,1 \\"},{"line_number":87,"context_line":"       --property hw:numa_cpus.1\u003d2,3,4,5,6,7 \\"},{"line_number":88,"context_line":"       --property hw:numa_mem.0\u003d2 \\"},{"line_number":89,"context_line":"       --property hw:numa_mem.1\u003d6"},{"line_number":90,"context_line":""},{"line_number":91,"context_line":"The NUMA topology exposed inside the guest can be overriden. To configure a"},{"line_number":92,"context_line":"flavor with eight cores to provide a two socket, two core and two thread"}],"source_content_type":"text/x-rst","patch_set":3,"id":"dab17558_819e6654","line":89,"in_reply_to":"dab17558_edbacabf","updated":"2016-05-06 15:18:44.000000000","message":"Way ahead of you :) \n\n    https://review.openstack.org/#/c/313538/1\n\nI haven\u0027t linked it in here but will do so once I\u0027ve the pinning section done.","commit_id":"d9cad91e3c20e79da095040dc1eeff180b2d172b"},{"author":{"_account_id":19779,"name":"Chason Chan","display_name":"chenxing","email":"chason.chan@foxmail.com","username":"chenxing"},"change_message_id":"a0016f5b8128bb019780c903019df443c2a4709c","unresolved":false,"context_lines":[{"line_number":75,"context_line":"NUMA node. For such workloads, it is beneficial to spread the instance across"},{"line_number":76,"context_line":"multiple host NUMA nodes, even if the instance\u0027s RAM/vCPUs could theoretically"},{"line_number":77,"context_line":"fit on a single NUMA node. To force an instance\u0027s VCPUs to spread across two"},{"line_number":78,"context_line":"host NUMA nodes, execute:"},{"line_number":79,"context_line":""},{"line_number":80,"context_line":".. code-block:: console"},{"line_number":81,"context_line":""}],"source_content_type":"text/x-rst","patch_set":5,"id":"dab17558_a91b751a","line":78,"range":{"start_line":78,"start_character":17,"end_line":78,"end_character":24},"updated":"2016-05-07 13:16:51.000000000","message":"s/execute/run/g","commit_id":"e19051fd112f1f9855194c8d2f2362a2915266e4"},{"author":{"_account_id":15334,"name":"Stephen Finucane","display_name":"stephenfin","email":"stephenfin@redhat.com","username":"sfinucan"},"change_message_id":"9a7baa523d702482fa6d659fbdb49193d86fc5bf","unresolved":false,"context_lines":[{"line_number":75,"context_line":"NUMA node. For such workloads, it is beneficial to spread the instance across"},{"line_number":76,"context_line":"multiple host NUMA nodes, even if the instance\u0027s RAM/vCPUs could theoretically"},{"line_number":77,"context_line":"fit on a single NUMA node. To force an instance\u0027s VCPUs to spread across two"},{"line_number":78,"context_line":"host NUMA nodes, execute:"},{"line_number":79,"context_line":""},{"line_number":80,"context_line":".. code-block:: console"},{"line_number":81,"context_line":""}],"source_content_type":"text/x-rst","patch_set":5,"id":"dab17558_d6b63fdb","line":78,"range":{"start_line":78,"start_character":17,"end_line":78,"end_character":24},"in_reply_to":"dab17558_a91b751a","updated":"2016-05-09 11:11:52.000000000","message":"Done","commit_id":"e19051fd112f1f9855194c8d2f2362a2915266e4"},{"author":{"_account_id":19779,"name":"Chason Chan","display_name":"chenxing","email":"chason.chan@foxmail.com","username":"chenxing"},"change_message_id":"a0016f5b8128bb019780c903019df443c2a4709c","unresolved":false,"context_lines":[{"line_number":85,"context_line":"configured. This allows for asymmetric allocation of CPUs and memory, which"},{"line_number":86,"context_line":"can be important for some workloads. To spread an instance\u0027s vCPUs across two"},{"line_number":87,"context_line":"NUMA nodes and create an asymmetric 1:2 vCPU and memory mappings betwen the"},{"line_number":88,"context_line":"two nodes, execute:"},{"line_number":89,"context_line":""},{"line_number":90,"context_line":".. code-block:: console"},{"line_number":91,"context_line":""}],"source_content_type":"text/x-rst","patch_set":5,"id":"dab17558_891671ef","line":88,"range":{"start_line":88,"start_character":11,"end_line":88,"end_character":18},"updated":"2016-05-07 13:16:51.000000000","message":"ditto","commit_id":"e19051fd112f1f9855194c8d2f2362a2915266e4"},{"author":{"_account_id":15334,"name":"Stephen Finucane","display_name":"stephenfin","email":"stephenfin@redhat.com","username":"sfinucan"},"change_message_id":"9a7baa523d702482fa6d659fbdb49193d86fc5bf","unresolved":false,"context_lines":[{"line_number":85,"context_line":"configured. This allows for asymmetric allocation of CPUs and memory, which"},{"line_number":86,"context_line":"can be important for some workloads. To spread an instance\u0027s vCPUs across two"},{"line_number":87,"context_line":"NUMA nodes and create an asymmetric 1:2 vCPU and memory mappings betwen the"},{"line_number":88,"context_line":"two nodes, execute:"},{"line_number":89,"context_line":""},{"line_number":90,"context_line":".. code-block:: console"},{"line_number":91,"context_line":""}],"source_content_type":"text/x-rst","patch_set":5,"id":"dab17558_b6b9fbcd","line":88,"range":{"start_line":88,"start_character":11,"end_line":88,"end_character":18},"in_reply_to":"dab17558_891671ef","updated":"2016-05-09 11:11:52.000000000","message":"Done","commit_id":"e19051fd112f1f9855194c8d2f2362a2915266e4"},{"author":{"_account_id":6772,"name":"Stephen Gordon","email":"sgordon@redhat.com","username":"sgordon"},"change_message_id":"0fa506bb3b8a63916bee29639fb293c270714982","unresolved":false,"context_lines":[{"line_number":100,"context_line":""},{"line_number":101,"context_line":"   The NUMA topology shown in the instance will not necessarily reflect the"},{"line_number":102,"context_line":"   underlying topology of the instance. This is by design. See this"},{"line_number":103,"context_line":"   `bug \u003chttps://bugs.launchpad.net/nova/+bug/1466780\u003e`_ for more information."},{"line_number":104,"context_line":""},{"line_number":105,"context_line":"For more information about the syntax for ``hw:numa_nodes``,"},{"line_number":106,"context_line":"``hw:numa_cpus.N`` and ``hw:num_mem.N``, refer to the `flavors guide`_."}],"source_content_type":"text/x-rst","patch_set":5,"id":"dab17558_26de8711","line":103,"updated":"2016-05-08 23:02:59.000000000","message":"Here you are referring to the topology of the object correct? Most users I think would expect you are talking about the topology reflected in the Libvirt XML which if I understand correctly is right? Perhaps best to clarify.","commit_id":"e19051fd112f1f9855194c8d2f2362a2915266e4"},{"author":{"_account_id":15334,"name":"Stephen Finucane","display_name":"stephenfin","email":"stephenfin@redhat.com","username":"sfinucan"},"change_message_id":"9a7baa523d702482fa6d659fbdb49193d86fc5bf","unresolved":false,"context_lines":[{"line_number":100,"context_line":""},{"line_number":101,"context_line":"   The NUMA topology shown in the instance will not necessarily reflect the"},{"line_number":102,"context_line":"   underlying topology of the instance. This is by design. See this"},{"line_number":103,"context_line":"   `bug \u003chttps://bugs.launchpad.net/nova/+bug/1466780\u003e`_ for more information."},{"line_number":104,"context_line":""},{"line_number":105,"context_line":"For more information about the syntax for ``hw:numa_nodes``,"},{"line_number":106,"context_line":"``hw:numa_cpus.N`` and ``hw:num_mem.N``, refer to the `flavors guide`_."}],"source_content_type":"text/x-rst","patch_set":5,"id":"dab17558_56a40fa5","line":103,"in_reply_to":"dab17558_26de8711","updated":"2016-05-09 11:11:52.000000000","message":"I\u0027ve clarified this a bit and moved it to the pinning section as this is where it applies (when NUMA topology and pinning is combined). Let me know what you think?","commit_id":"e19051fd112f1f9855194c8d2f2362a2915266e4"},{"author":{"_account_id":19779,"name":"Chason Chan","display_name":"chenxing","email":"chason.chan@foxmail.com","username":"chenxing"},"change_message_id":"a0016f5b8128bb019780c903019df443c2a4709c","unresolved":false,"context_lines":[{"line_number":136,"context_line":"of components and contention on these components can impact performance. To"},{"line_number":137,"context_line":"configure how threads should be used, a CPU thread policy should be specified."},{"line_number":138,"context_line":"For workloads where sharing benefits performance, thread siblings should be"},{"line_number":139,"context_line":"used. To force this, execute:"},{"line_number":140,"context_line":""},{"line_number":141,"context_line":".. code-block:: console"},{"line_number":142,"context_line":""}],"source_content_type":"text/x-rst","patch_set":5,"id":"dab17558_29278565","line":139,"range":{"start_line":139,"start_character":21,"end_line":139,"end_character":28},"updated":"2016-05-07 13:16:51.000000000","message":"ditto","commit_id":"e19051fd112f1f9855194c8d2f2362a2915266e4"},{"author":{"_account_id":15334,"name":"Stephen Finucane","display_name":"stephenfin","email":"stephenfin@redhat.com","username":"sfinucan"},"change_message_id":"9a7baa523d702482fa6d659fbdb49193d86fc5bf","unresolved":false,"context_lines":[{"line_number":136,"context_line":"of components and contention on these components can impact performance. To"},{"line_number":137,"context_line":"configure how threads should be used, a CPU thread policy should be specified."},{"line_number":138,"context_line":"For workloads where sharing benefits performance, thread siblings should be"},{"line_number":139,"context_line":"used. To force this, execute:"},{"line_number":140,"context_line":""},{"line_number":141,"context_line":".. code-block:: console"},{"line_number":142,"context_line":""}],"source_content_type":"text/x-rst","patch_set":5,"id":"dab17558_16b1e7e1","line":139,"range":{"start_line":139,"start_character":21,"end_line":139,"end_character":28},"in_reply_to":"dab17558_29278565","updated":"2016-05-09 11:11:52.000000000","message":"Done","commit_id":"e19051fd112f1f9855194c8d2f2362a2915266e4"},{"author":{"_account_id":19779,"name":"Chason Chan","display_name":"chenxing","email":"chason.chan@foxmail.com","username":"chenxing"},"change_message_id":"a0016f5b8128bb019780c903019df443c2a4709c","unresolved":false,"context_lines":[{"line_number":145,"context_line":"       --property hw:cpu_thread_policy\u003drequire"},{"line_number":146,"context_line":""},{"line_number":147,"context_line":"For other workloads where performance is impacted by contention for resources,"},{"line_number":148,"context_line":"non-thread siblings or non-SMT hosts should be used. To force this, execute:"},{"line_number":149,"context_line":""},{"line_number":150,"context_line":".. code-block:: console"},{"line_number":151,"context_line":""}],"source_content_type":"text/x-rst","patch_set":5,"id":"dab17558_c9d6b926","line":148,"range":{"start_line":148,"start_character":68,"end_line":148,"end_character":75},"updated":"2016-05-07 13:16:51.000000000","message":"ditto","commit_id":"e19051fd112f1f9855194c8d2f2362a2915266e4"},{"author":{"_account_id":15334,"name":"Stephen Finucane","display_name":"stephenfin","email":"stephenfin@redhat.com","username":"sfinucan"},"change_message_id":"9a7baa523d702482fa6d659fbdb49193d86fc5bf","unresolved":false,"context_lines":[{"line_number":145,"context_line":"       --property hw:cpu_thread_policy\u003drequire"},{"line_number":146,"context_line":""},{"line_number":147,"context_line":"For other workloads where performance is impacted by contention for resources,"},{"line_number":148,"context_line":"non-thread siblings or non-SMT hosts should be used. To force this, execute:"},{"line_number":149,"context_line":""},{"line_number":150,"context_line":".. code-block:: console"},{"line_number":151,"context_line":""}],"source_content_type":"text/x-rst","patch_set":5,"id":"dab17558_f6b383e7","line":148,"range":{"start_line":148,"start_character":68,"end_line":148,"end_character":75},"in_reply_to":"dab17558_c9d6b926","updated":"2016-05-09 11:11:52.000000000","message":"Done","commit_id":"e19051fd112f1f9855194c8d2f2362a2915266e4"},{"author":{"_account_id":17973,"name":"Ludovic Beliveau","email":"lbeliveau@gmail.com","username":"lbeliveau"},"change_message_id":"c888d4dbf4ed9f1948707ba638d20a460e675ec2","unresolved":false,"context_lines":[{"line_number":42,"context_line":"SMP systems share a bus and some memory, CPUs in SMT systems share many more"},{"line_number":43,"context_line":"components. CPUs that share components are known as thread siblings. All cores"},{"line_number":44,"context_line":"appear as usable cores on the system; however, as with NUMA competition for"},{"line_number":45,"context_line":"components shared between thread siblings can cause performance penalties."},{"line_number":46,"context_line":""},{"line_number":47,"context_line":"Customing instance NUMA topologies"},{"line_number":48,"context_line":"~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~"}],"source_content_type":"text/x-rst","patch_set":6,"id":"dab17558_cfddfe15","line":45,"updated":"2016-05-09 12:45:30.000000000","message":"It\u0027s not clear in this paragraph that there are multiple hardware threads that can execute in parallel on the same core.","commit_id":"dfa2f357d84366444c4b49005e2b5b9ea5420175"},{"author":{"_account_id":17973,"name":"Ludovic Beliveau","email":"lbeliveau@gmail.com","username":"lbeliveau"},"change_message_id":"c888d4dbf4ed9f1948707ba638d20a460e675ec2","unresolved":false,"context_lines":[{"line_number":96,"context_line":"       --property hw:numa_cpus.0\u003d0,1 \\"},{"line_number":97,"context_line":"       --property hw:numa_cpus.1\u003d2,3,4,5 \\"},{"line_number":98,"context_line":"       --property hw:numa_mem.0\u003d2048 \\"},{"line_number":99,"context_line":"       --property hw:numa_mem.1\u003d4096"},{"line_number":100,"context_line":""},{"line_number":101,"context_line":"For more information about the syntax for ``hw:numa_nodes``,"},{"line_number":102,"context_line":"``hw:numa_cpus.N`` and ``hw:num_mem.N``, refer to the `flavors guide`_."}],"source_content_type":"text/x-rst","patch_set":6,"id":"dab17558_2f692a26","line":99,"updated":"2016-05-09 12:45:30.000000000","message":"Even if the syntax is explained in the flavor, I would still explain in more detail what this command do.  This would give more clarity to the reader.","commit_id":"dfa2f357d84366444c4b49005e2b5b9ea5420175"},{"author":{"_account_id":17973,"name":"Ludovic Beliveau","email":"lbeliveau@gmail.com","username":"lbeliveau"},"change_message_id":"47291e8ccfc8bf0e79adef2c7d968bb6a94836cf","unresolved":false,"context_lines":[{"line_number":41,"context_line":"Intel platforms, is an design that is complementary to SMP. Whereas CPUs in"},{"line_number":42,"context_line":"SMP systems share a bus and some memory, CPUs in SMT systems share many more"},{"line_number":43,"context_line":"components. CPUs that share components are known as thread siblings. All CPUs"},{"line_number":44,"context_line":"appear as usable CPUs on the system and can execute workloads in parallel;"},{"line_number":45,"context_line":"however, as with NUMA competition for components shared between thread"},{"line_number":46,"context_line":"siblings can cause performance penalties."},{"line_number":47,"context_line":""}],"source_content_type":"text/x-rst","patch_set":7,"id":"dab17558_0e042348","line":44,"updated":"2016-05-09 14:51:34.000000000","message":"The last part need re-phrasing:\n\nas with NUMA, threads compete for shared components.","commit_id":"e32d87b873a8811fb3b88d82bb0962c3898033e3"},{"author":{"_account_id":15334,"name":"Stephen Finucane","display_name":"stephenfin","email":"stephenfin@redhat.com","username":"sfinucan"},"change_message_id":"2a18e27dc671c6a8b32922164af2a750c834427f","unresolved":false,"context_lines":[{"line_number":41,"context_line":"Intel platforms, is an design that is complementary to SMP. Whereas CPUs in"},{"line_number":42,"context_line":"SMP systems share a bus and some memory, CPUs in SMT systems share many more"},{"line_number":43,"context_line":"components. CPUs that share components are known as thread siblings. All CPUs"},{"line_number":44,"context_line":"appear as usable CPUs on the system and can execute workloads in parallel;"},{"line_number":45,"context_line":"however, as with NUMA competition for components shared between thread"},{"line_number":46,"context_line":"siblings can cause performance penalties."},{"line_number":47,"context_line":""}],"source_content_type":"text/x-rst","patch_set":7,"id":"dab17558_b413aa0c","line":44,"in_reply_to":"dab17558_0e042348","updated":"2016-05-09 15:14:41.000000000","message":"Done","commit_id":"e32d87b873a8811fb3b88d82bb0962c3898033e3"},{"author":{"_account_id":17973,"name":"Ludovic Beliveau","email":"lbeliveau@gmail.com","username":"lbeliveau"},"change_message_id":"47291e8ccfc8bf0e79adef2c7d968bb6a94836cf","unresolved":false,"context_lines":[{"line_number":93,"context_line":".. code-block:: console"},{"line_number":94,"context_line":""},{"line_number":95,"context_line":"   # openstack flavor set m1.large --property hw:numa_nodes\u003d2"},{"line_number":96,"context_line":"   # openstack flavor set m1.large \\  # configure the first guest node"},{"line_number":97,"context_line":"       --property hw:numa_cpus.0\u003d0,1 \\"},{"line_number":98,"context_line":"       --property hw:numa_mem.0\u003d2048"},{"line_number":99,"context_line":"   # openstack flavor set m1.large \\  # configure the second guest node"}],"source_content_type":"text/x-rst","patch_set":7,"id":"dab17558_6e67af25","line":96,"updated":"2016-05-09 14:51:34.000000000","message":"Maybe explain what 0 and 1 are here.","commit_id":"e32d87b873a8811fb3b88d82bb0962c3898033e3"},{"author":{"_account_id":15334,"name":"Stephen Finucane","display_name":"stephenfin","email":"stephenfin@redhat.com","username":"sfinucan"},"change_message_id":"2a18e27dc671c6a8b32922164af2a750c834427f","unresolved":false,"context_lines":[{"line_number":93,"context_line":".. code-block:: console"},{"line_number":94,"context_line":""},{"line_number":95,"context_line":"   # openstack flavor set m1.large --property hw:numa_nodes\u003d2"},{"line_number":96,"context_line":"   # openstack flavor set m1.large \\  # configure the first guest node"},{"line_number":97,"context_line":"       --property hw:numa_cpus.0\u003d0,1 \\"},{"line_number":98,"context_line":"       --property hw:numa_mem.0\u003d2048"},{"line_number":99,"context_line":"   # openstack flavor set m1.large \\  # configure the second guest node"}],"source_content_type":"text/x-rst","patch_set":7,"id":"dab17558_54051eba","line":96,"in_reply_to":"dab17558_6e67af25","updated":"2016-05-09 15:14:41.000000000","message":"Done","commit_id":"e32d87b873a8811fb3b88d82bb0962c3898033e3"}],"doc/admin-guide/source/compute-numa-cpu-pinning.rst":[{"author":{"_account_id":17973,"name":"Ludovic Beliveau","email":"lbeliveau@gmail.com","username":"lbeliveau"},"change_message_id":"ae6714dd709857ddb3a4b2cbed43f6d6cc47ad7e","unresolved":false,"context_lines":[{"line_number":29,"context_line":"systems share a bus and some memory, CPUs in SMT systems share many more"},{"line_number":30,"context_line":"components. CPUs that share components are known as thread siblings. All CPUs"},{"line_number":31,"context_line":"appear as usable CPUs on the system and can execute workloads in parallel;"},{"line_number":32,"context_line":"however, as with NUMA, threads for compete for shared resources."},{"line_number":33,"context_line":""},{"line_number":34,"context_line":"Customing instance NUMA topologies"},{"line_number":35,"context_line":"~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~"}],"source_content_type":"text/x-rst","patch_set":9,"id":"dab17558_04699f8b","line":32,"updated":"2016-05-09 16:54:46.000000000","message":"d/for/","commit_id":"4575d652270f0113993c209202f640ec9ddb3515"},{"author":{"_account_id":15334,"name":"Stephen Finucane","display_name":"stephenfin","email":"stephenfin@redhat.com","username":"sfinucan"},"change_message_id":"b33b7a14db9ab9b410bf93c28a2f13351f2b1fc2","unresolved":false,"context_lines":[{"line_number":29,"context_line":"systems share a bus and some memory, CPUs in SMT systems share many more"},{"line_number":30,"context_line":"components. CPUs that share components are known as thread siblings. All CPUs"},{"line_number":31,"context_line":"appear as usable CPUs on the system and can execute workloads in parallel;"},{"line_number":32,"context_line":"however, as with NUMA, threads for compete for shared resources."},{"line_number":33,"context_line":""},{"line_number":34,"context_line":"Customing instance NUMA topologies"},{"line_number":35,"context_line":"~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~"}],"source_content_type":"text/x-rst","patch_set":9,"id":"dab17558_c747991b","line":32,"in_reply_to":"dab17558_04699f8b","updated":"2016-05-09 17:20:31.000000000","message":"Done","commit_id":"4575d652270f0113993c209202f640ec9ddb3515"},{"author":{"_account_id":17973,"name":"Ludovic Beliveau","email":"lbeliveau@gmail.com","username":"lbeliveau"},"change_message_id":"083c4313f7f148b83d24954f604a115ebdcef9a6","unresolved":false,"context_lines":[{"line_number":29,"context_line":"systems share a bus and some memory, CPUs in SMT systems share many more"},{"line_number":30,"context_line":"components. CPUs that share components are known as thread siblings. All CPUs"},{"line_number":31,"context_line":"appear as usable CPUs on the system and can execute workloads in parallel;"},{"line_number":32,"context_line":"however, as with NUMA, threads compete for shared resources."},{"line_number":33,"context_line":""},{"line_number":34,"context_line":"Customing instance NUMA topologies"},{"line_number":35,"context_line":"~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~"}],"source_content_type":"text/x-rst","patch_set":10,"id":"dab17558_c71cb915","line":32,"updated":"2016-05-09 17:23:37.000000000","message":"Nit, up to you, but I would maybe just break this last sentence in two.  But not necessary ...","commit_id":"4780ccab75d3e41f0657e04829a30ce1b44fa997"},{"author":{"_account_id":17207,"name":"Atsushi SAKAI","email":"sakaia@jp.fujitsu.com","username":"sakaia"},"change_message_id":"f69af81e03e3f30be358cea47f3febe954e0262f","unresolved":false,"context_lines":[{"line_number":31,"context_line":"appear as usable CPUs on the system and can execute workloads in parallel;"},{"line_number":32,"context_line":"however, as with NUMA, threads compete for shared resources."},{"line_number":33,"context_line":""},{"line_number":34,"context_line":"Customing instance NUMA topologies"},{"line_number":35,"context_line":"~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~"},{"line_number":36,"context_line":""},{"line_number":37,"context_line":".. note::"}],"source_content_type":"text/x-rst","patch_set":11,"id":"dab17558_bc750c99","line":34,"updated":"2016-05-10 17:02:13.000000000","message":"Customing \u003d\u003e Customizing\n\nAlso underline length should fix accoringly.","commit_id":"8038ad370b479e081d7deef06945de919a9e0285"},{"author":{"_account_id":15334,"name":"Stephen Finucane","display_name":"stephenfin","email":"stephenfin@redhat.com","username":"sfinucan"},"change_message_id":"48b54951056835a7fc2bea8bedffd8ed392330a2","unresolved":false,"context_lines":[{"line_number":31,"context_line":"appear as usable CPUs on the system and can execute workloads in parallel;"},{"line_number":32,"context_line":"however, as with NUMA, threads compete for shared resources."},{"line_number":33,"context_line":""},{"line_number":34,"context_line":"Customing instance NUMA topologies"},{"line_number":35,"context_line":"~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~"},{"line_number":36,"context_line":""},{"line_number":37,"context_line":".. note::"}],"source_content_type":"text/x-rst","patch_set":11,"id":"dab17558_05a2dd98","line":34,"in_reply_to":"dab17558_bc750c99","updated":"2016-05-11 09:07:02.000000000","message":"Done","commit_id":"8038ad370b479e081d7deef06945de919a9e0285"},{"author":{"_account_id":17207,"name":"Atsushi SAKAI","email":"sakaia@jp.fujitsu.com","username":"sakaia"},"change_message_id":"f69af81e03e3f30be358cea47f3febe954e0262f","unresolved":false,"context_lines":[{"line_number":138,"context_line":"       --property hw:cpu_thread_policy\u003dseparate"},{"line_number":139,"context_line":""},{"line_number":140,"context_line":"Finally for workloads where performance is minimally impacted, thread siblings"},{"line_number":141,"context_line":"can be prefered. This is the default, but it can be set explicitly:"},{"line_number":142,"context_line":""},{"line_number":143,"context_line":".. code-block:: console"},{"line_number":144,"context_line":""}],"source_content_type":"text/x-rst","patch_set":11,"id":"dab17558_a9cdcc32","line":141,"updated":"2016-05-10 17:02:13.000000000","message":"prefered \u003d\u003e preferred","commit_id":"8038ad370b479e081d7deef06945de919a9e0285"},{"author":{"_account_id":15334,"name":"Stephen Finucane","display_name":"stephenfin","email":"stephenfin@redhat.com","username":"sfinucan"},"change_message_id":"48b54951056835a7fc2bea8bedffd8ed392330a2","unresolved":false,"context_lines":[{"line_number":138,"context_line":"       --property hw:cpu_thread_policy\u003dseparate"},{"line_number":139,"context_line":""},{"line_number":140,"context_line":"Finally for workloads where performance is minimally impacted, thread siblings"},{"line_number":141,"context_line":"can be prefered. This is the default, but it can be set explicitly:"},{"line_number":142,"context_line":""},{"line_number":143,"context_line":".. code-block:: console"},{"line_number":144,"context_line":""}],"source_content_type":"text/x-rst","patch_set":11,"id":"dab17558_2543a1b3","line":141,"in_reply_to":"dab17558_a9cdcc32","updated":"2016-05-11 09:07:02.000000000","message":"Done","commit_id":"8038ad370b479e081d7deef06945de919a9e0285"},{"author":{"_account_id":17207,"name":"Atsushi SAKAI","email":"sakaia@jp.fujitsu.com","username":"sakaia"},"change_message_id":"f69af81e03e3f30be358cea47f3febe954e0262f","unresolved":false,"context_lines":[{"line_number":148,"context_line":""},{"line_number":149,"context_line":".. note::"},{"line_number":150,"context_line":""},{"line_number":151,"context_line":"   There is no correlation required between the NUMA toplogy exposed in the"},{"line_number":152,"context_line":"   instance and how the instance is actually pinned on the host. This is by"},{"line_number":153,"context_line":"   design. See this `bug \u003chttps://bugs.launchpad.net/nova/+bug/1466780\u003e`_ for"},{"line_number":154,"context_line":"   more information."}],"source_content_type":"text/x-rst","patch_set":11,"id":"dab17558_a9666c4e","line":151,"updated":"2016-05-10 17:02:13.000000000","message":"toplogy \u003d\u003e topology","commit_id":"8038ad370b479e081d7deef06945de919a9e0285"},{"author":{"_account_id":15334,"name":"Stephen Finucane","display_name":"stephenfin","email":"stephenfin@redhat.com","username":"sfinucan"},"change_message_id":"48b54951056835a7fc2bea8bedffd8ed392330a2","unresolved":false,"context_lines":[{"line_number":148,"context_line":""},{"line_number":149,"context_line":".. note::"},{"line_number":150,"context_line":""},{"line_number":151,"context_line":"   There is no correlation required between the NUMA toplogy exposed in the"},{"line_number":152,"context_line":"   instance and how the instance is actually pinned on the host. This is by"},{"line_number":153,"context_line":"   design. See this `bug \u003chttps://bugs.launchpad.net/nova/+bug/1466780\u003e`_ for"},{"line_number":154,"context_line":"   more information."}],"source_content_type":"text/x-rst","patch_set":11,"id":"dab17558_053e5d27","line":151,"in_reply_to":"dab17558_a9666c4e","updated":"2016-05-11 09:07:02.000000000","message":"Done","commit_id":"8038ad370b479e081d7deef06945de919a9e0285"},{"author":{"_account_id":14962,"name":"ologvinova","email":"ologvinova@mirantis.com","username":"ologvinova"},"change_message_id":"081785ae28507f3a6a2992fb122589990023f088","unresolved":false,"context_lines":[{"line_number":9,"context_line":"inside the instance. These features can be used to minimize latency and"},{"line_number":10,"context_line":"maximize per-instance performance."},{"line_number":11,"context_line":""},{"line_number":12,"context_line":"SMP, NUMA and SMT overviews"},{"line_number":13,"context_line":"~~~~~~~~~~~~~~~~~~~~~~~~~~~"},{"line_number":14,"context_line":""},{"line_number":15,"context_line":"Symmetric multiprocessing (SMP) is a design found in many modern,"}],"source_content_type":"text/x-rst","patch_set":12,"id":"dab17558_9b8bce1e","line":12,"range":{"start_line":12,"start_character":9,"end_line":12,"end_character":10},"updated":"2016-05-11 11:35:12.000000000","message":"comma","commit_id":"2626a08c524ec267ced6b55cca0c05289ac22510"},{"author":{"_account_id":15334,"name":"Stephen Finucane","display_name":"stephenfin","email":"stephenfin@redhat.com","username":"sfinucan"},"change_message_id":"0c2e44efc3f1d7f40d92f82d1eb3d115935763a2","unresolved":false,"context_lines":[{"line_number":9,"context_line":"inside the instance. These features can be used to minimize latency and"},{"line_number":10,"context_line":"maximize per-instance performance."},{"line_number":11,"context_line":""},{"line_number":12,"context_line":"SMP, NUMA and SMT overviews"},{"line_number":13,"context_line":"~~~~~~~~~~~~~~~~~~~~~~~~~~~"},{"line_number":14,"context_line":""},{"line_number":15,"context_line":"Symmetric multiprocessing (SMP) is a design found in many modern,"}],"source_content_type":"text/x-rst","patch_set":12,"id":"dab17558_188649db","line":12,"range":{"start_line":12,"start_character":9,"end_line":12,"end_character":10},"in_reply_to":"dab17558_9b8bce1e","updated":"2016-05-13 09:50:19.000000000","message":"Done","commit_id":"2626a08c524ec267ced6b55cca0c05289ac22510"},{"author":{"_account_id":14962,"name":"ologvinova","email":"ologvinova@mirantis.com","username":"ologvinova"},"change_message_id":"081785ae28507f3a6a2992fb122589990023f088","unresolved":false,"context_lines":[{"line_number":12,"context_line":"SMP, NUMA and SMT overviews"},{"line_number":13,"context_line":"~~~~~~~~~~~~~~~~~~~~~~~~~~~"},{"line_number":14,"context_line":""},{"line_number":15,"context_line":"Symmetric multiprocessing (SMP) is a design found in many modern,"},{"line_number":16,"context_line":"multi-core systems. In an SMP system, there are two or more CPUs and these"},{"line_number":17,"context_line":"CPUs are connected by some interconnect. This provides CPUs with equal access"},{"line_number":18,"context_line":"to system resources like memory, IO ports etc."}],"source_content_type":"text/x-rst","patch_set":12,"id":"dab17558_9b92ae4a","line":15,"range":{"start_line":15,"start_character":64,"end_line":15,"end_character":65},"updated":"2016-05-11 11:35:12.000000000","message":"I would remove comma here.","commit_id":"2626a08c524ec267ced6b55cca0c05289ac22510"},{"author":{"_account_id":15334,"name":"Stephen Finucane","display_name":"stephenfin","email":"stephenfin@redhat.com","username":"sfinucan"},"change_message_id":"0c2e44efc3f1d7f40d92f82d1eb3d115935763a2","unresolved":false,"context_lines":[{"line_number":12,"context_line":"SMP, NUMA and SMT overviews"},{"line_number":13,"context_line":"~~~~~~~~~~~~~~~~~~~~~~~~~~~"},{"line_number":14,"context_line":""},{"line_number":15,"context_line":"Symmetric multiprocessing (SMP) is a design found in many modern,"},{"line_number":16,"context_line":"multi-core systems. In an SMP system, there are two or more CPUs and these"},{"line_number":17,"context_line":"CPUs are connected by some interconnect. This provides CPUs with equal access"},{"line_number":18,"context_line":"to system resources like memory, IO ports etc."}],"source_content_type":"text/x-rst","patch_set":12,"id":"dab17558_f8dc251a","line":15,"range":{"start_line":15,"start_character":64,"end_line":15,"end_character":65},"in_reply_to":"dab17558_9b92ae4a","updated":"2016-05-13 09:50:19.000000000","message":"Done","commit_id":"2626a08c524ec267ced6b55cca0c05289ac22510"},{"author":{"_account_id":14962,"name":"ologvinova","email":"ologvinova@mirantis.com","username":"ologvinova"},"change_message_id":"081785ae28507f3a6a2992fb122589990023f088","unresolved":false,"context_lines":[{"line_number":15,"context_line":"Symmetric multiprocessing (SMP) is a design found in many modern,"},{"line_number":16,"context_line":"multi-core systems. In an SMP system, there are two or more CPUs and these"},{"line_number":17,"context_line":"CPUs are connected by some interconnect. This provides CPUs with equal access"},{"line_number":18,"context_line":"to system resources like memory, IO ports etc."},{"line_number":19,"context_line":""},{"line_number":20,"context_line":"Non-uniform memory access (NUMA) is a derivative of the SMP design that is"},{"line_number":21,"context_line":"found in many multi-socket systems. In a NUMA system, system memory is divided"}],"source_content_type":"text/x-rst","patch_set":12,"id":"dab17558_fba13a95","line":18,"range":{"start_line":18,"start_character":41,"end_line":18,"end_character":45},"updated":"2016-05-11 11:35:12.000000000","message":", and so on.","commit_id":"2626a08c524ec267ced6b55cca0c05289ac22510"},{"author":{"_account_id":15334,"name":"Stephen Finucane","display_name":"stephenfin","email":"stephenfin@redhat.com","username":"sfinucan"},"change_message_id":"0c2e44efc3f1d7f40d92f82d1eb3d115935763a2","unresolved":false,"context_lines":[{"line_number":15,"context_line":"Symmetric multiprocessing (SMP) is a design found in many modern,"},{"line_number":16,"context_line":"multi-core systems. In an SMP system, there are two or more CPUs and these"},{"line_number":17,"context_line":"CPUs are connected by some interconnect. This provides CPUs with equal access"},{"line_number":18,"context_line":"to system resources like memory, IO ports etc."},{"line_number":19,"context_line":""},{"line_number":20,"context_line":"Non-uniform memory access (NUMA) is a derivative of the SMP design that is"},{"line_number":21,"context_line":"found in many multi-socket systems. In a NUMA system, system memory is divided"}],"source_content_type":"text/x-rst","patch_set":12,"id":"dab17558_58cc5161","line":18,"range":{"start_line":18,"start_character":41,"end_line":18,"end_character":45},"in_reply_to":"dab17558_fba13a95","updated":"2016-05-13 09:50:19.000000000","message":"Done","commit_id":"2626a08c524ec267ced6b55cca0c05289ac22510"},{"author":{"_account_id":14962,"name":"ologvinova","email":"ologvinova@mirantis.com","username":"ologvinova"},"change_message_id":"081785ae28507f3a6a2992fb122589990023f088","unresolved":false,"context_lines":[{"line_number":21,"context_line":"found in many multi-socket systems. In a NUMA system, system memory is divided"},{"line_number":22,"context_line":"into cells or nodes that are associated with particular CPUs. Requests for"},{"line_number":23,"context_line":"memory on other nodes are possible through an interconnect bus; however,"},{"line_number":24,"context_line":"bandwidth across this shared bus is  limited and therefore competition for"},{"line_number":25,"context_line":"this resource can incur performance penalties."},{"line_number":26,"context_line":""},{"line_number":27,"context_line":"Simultaneous Multi-Threading (SMT), known as as Hyperthreading on Intel"}],"source_content_type":"text/x-rst","patch_set":12,"id":"dab17558_366c79cd","line":24,"range":{"start_line":24,"start_character":49,"end_line":24,"end_character":58},"updated":"2016-05-11 11:35:12.000000000","message":", therefore,","commit_id":"2626a08c524ec267ced6b55cca0c05289ac22510"},{"author":{"_account_id":15334,"name":"Stephen Finucane","display_name":"stephenfin","email":"stephenfin@redhat.com","username":"sfinucan"},"change_message_id":"0c2e44efc3f1d7f40d92f82d1eb3d115935763a2","unresolved":false,"context_lines":[{"line_number":21,"context_line":"found in many multi-socket systems. In a NUMA system, system memory is divided"},{"line_number":22,"context_line":"into cells or nodes that are associated with particular CPUs. Requests for"},{"line_number":23,"context_line":"memory on other nodes are possible through an interconnect bus; however,"},{"line_number":24,"context_line":"bandwidth across this shared bus is  limited and therefore competition for"},{"line_number":25,"context_line":"this resource can incur performance penalties."},{"line_number":26,"context_line":""},{"line_number":27,"context_line":"Simultaneous Multi-Threading (SMT), known as as Hyperthreading on Intel"}],"source_content_type":"text/x-rst","patch_set":12,"id":"dab17558_98d6b9ef","line":24,"range":{"start_line":24,"start_character":49,"end_line":24,"end_character":58},"in_reply_to":"dab17558_366c79cd","updated":"2016-05-13 09:50:19.000000000","message":"Done","commit_id":"2626a08c524ec267ced6b55cca0c05289ac22510"},{"author":{"_account_id":14962,"name":"ologvinova","email":"ologvinova@mirantis.com","username":"ologvinova"},"change_message_id":"081785ae28507f3a6a2992fb122589990023f088","unresolved":false,"context_lines":[{"line_number":24,"context_line":"bandwidth across this shared bus is  limited and therefore competition for"},{"line_number":25,"context_line":"this resource can incur performance penalties."},{"line_number":26,"context_line":""},{"line_number":27,"context_line":"Simultaneous Multi-Threading (SMT), known as as Hyperthreading on Intel"},{"line_number":28,"context_line":"platforms, is an design that is complementary to SMP. Whereas CPUs in SMP"},{"line_number":29,"context_line":"systems share a bus and some memory, CPUs in SMT systems share many more"},{"line_number":30,"context_line":"components. CPUs that share components are known as thread siblings. All CPUs"}],"source_content_type":"text/x-rst","patch_set":12,"id":"dab17558_f67b1179","line":27,"range":{"start_line":27,"start_character":48,"end_line":27,"end_character":62},"updated":"2016-05-11 11:35:12.000000000","message":"Hyper-Threading","commit_id":"2626a08c524ec267ced6b55cca0c05289ac22510"},{"author":{"_account_id":15334,"name":"Stephen Finucane","display_name":"stephenfin","email":"stephenfin@redhat.com","username":"sfinucan"},"change_message_id":"0c2e44efc3f1d7f40d92f82d1eb3d115935763a2","unresolved":false,"context_lines":[{"line_number":24,"context_line":"bandwidth across this shared bus is  limited and therefore competition for"},{"line_number":25,"context_line":"this resource can incur performance penalties."},{"line_number":26,"context_line":""},{"line_number":27,"context_line":"Simultaneous Multi-Threading (SMT), known as as Hyperthreading on Intel"},{"line_number":28,"context_line":"platforms, is an design that is complementary to SMP. Whereas CPUs in SMP"},{"line_number":29,"context_line":"systems share a bus and some memory, CPUs in SMT systems share many more"},{"line_number":30,"context_line":"components. CPUs that share components are known as thread siblings. All CPUs"}],"source_content_type":"text/x-rst","patch_set":12,"id":"dab17558_d814c1a0","line":27,"range":{"start_line":27,"start_character":48,"end_line":27,"end_character":62},"in_reply_to":"dab17558_f67b1179","updated":"2016-05-13 09:50:19.000000000","message":"This is consistent with other docs. I\u0027ve made this change, however, so they should probably be fixed now","commit_id":"2626a08c524ec267ced6b55cca0c05289ac22510"},{"author":{"_account_id":14962,"name":"ologvinova","email":"ologvinova@mirantis.com","username":"ologvinova"},"change_message_id":"081785ae28507f3a6a2992fb122589990023f088","unresolved":false,"context_lines":[{"line_number":25,"context_line":"this resource can incur performance penalties."},{"line_number":26,"context_line":""},{"line_number":27,"context_line":"Simultaneous Multi-Threading (SMT), known as as Hyperthreading on Intel"},{"line_number":28,"context_line":"platforms, is an design that is complementary to SMP. Whereas CPUs in SMP"},{"line_number":29,"context_line":"systems share a bus and some memory, CPUs in SMT systems share many more"},{"line_number":30,"context_line":"components. CPUs that share components are known as thread siblings. All CPUs"},{"line_number":31,"context_line":"appear as usable CPUs on the system and can execute workloads in parallel;"}],"source_content_type":"text/x-rst","patch_set":12,"id":"dab17558_36e1192e","line":28,"range":{"start_line":28,"start_character":14,"end_line":28,"end_character":16},"updated":"2016-05-11 11:35:12.000000000","message":"a","commit_id":"2626a08c524ec267ced6b55cca0c05289ac22510"},{"author":{"_account_id":15334,"name":"Stephen Finucane","display_name":"stephenfin","email":"stephenfin@redhat.com","username":"sfinucan"},"change_message_id":"0c2e44efc3f1d7f40d92f82d1eb3d115935763a2","unresolved":false,"context_lines":[{"line_number":25,"context_line":"this resource can incur performance penalties."},{"line_number":26,"context_line":""},{"line_number":27,"context_line":"Simultaneous Multi-Threading (SMT), known as as Hyperthreading on Intel"},{"line_number":28,"context_line":"platforms, is an design that is complementary to SMP. Whereas CPUs in SMP"},{"line_number":29,"context_line":"systems share a bus and some memory, CPUs in SMT systems share many more"},{"line_number":30,"context_line":"components. CPUs that share components are known as thread siblings. All CPUs"},{"line_number":31,"context_line":"appear as usable CPUs on the system and can execute workloads in parallel;"}],"source_content_type":"text/x-rst","patch_set":12,"id":"dab17558_180f6989","line":28,"range":{"start_line":28,"start_character":14,"end_line":28,"end_character":16},"in_reply_to":"dab17558_36e1192e","updated":"2016-05-13 09:50:19.000000000","message":"Done","commit_id":"2626a08c524ec267ced6b55cca0c05289ac22510"},{"author":{"_account_id":14962,"name":"ologvinova","email":"ologvinova@mirantis.com","username":"ologvinova"},"change_message_id":"081785ae28507f3a6a2992fb122589990023f088","unresolved":false,"context_lines":[{"line_number":72,"context_line":""},{"line_number":73,"context_line":"The allocation of instances vCPUs and memory from different host NUMA nodes"},{"line_number":74,"context_line":"can be configured. This allows for asymmetric allocation of vCPUs and memory,"},{"line_number":75,"context_line":"which can be important for some workloads. To spread the six vCPUs and 6GB of"},{"line_number":76,"context_line":"memory of an instance across two NUMA nodes and create an asymmetric 1:2 vCPU"},{"line_number":77,"context_line":"and memory mapping between the two nodes, run:"},{"line_number":78,"context_line":""}],"source_content_type":"text/x-rst","patch_set":12,"id":"dab17558_676f68fc","line":75,"range":{"start_line":75,"start_character":46,"end_line":75,"end_character":74},"updated":"2016-05-11 11:35:12.000000000","message":"spread six vCPUs and 6 GB","commit_id":"2626a08c524ec267ced6b55cca0c05289ac22510"},{"author":{"_account_id":14962,"name":"ologvinova","email":"ologvinova@mirantis.com","username":"ologvinova"},"change_message_id":"081785ae28507f3a6a2992fb122589990023f088","unresolved":false,"context_lines":[{"line_number":87,"context_line":"       --property hw:numa_mem.1\u003d4096"},{"line_number":88,"context_line":""},{"line_number":89,"context_line":"For more information about the syntax for ``hw:numa_nodes``,"},{"line_number":90,"context_line":"``hw:numa_cpus.N`` and ``hw:num_mem.N``, refer to the `flavors guide`_."},{"line_number":91,"context_line":""},{"line_number":92,"context_line":"Customizing instance CPU policies"},{"line_number":93,"context_line":"~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~"}],"source_content_type":"text/x-rst","patch_set":12,"id":"dab17558_8711ac5f","line":90,"range":{"start_line":90,"start_character":54,"end_line":90,"end_character":70},"updated":"2016-05-11 11:35:12.000000000","message":"`Flavors`_ guide","commit_id":"2626a08c524ec267ced6b55cca0c05289ac22510"},{"author":{"_account_id":15334,"name":"Stephen Finucane","display_name":"stephenfin","email":"stephenfin@redhat.com","username":"sfinucan"},"change_message_id":"0c2e44efc3f1d7f40d92f82d1eb3d115935763a2","unresolved":false,"context_lines":[{"line_number":87,"context_line":"       --property hw:numa_mem.1\u003d4096"},{"line_number":88,"context_line":""},{"line_number":89,"context_line":"For more information about the syntax for ``hw:numa_nodes``,"},{"line_number":90,"context_line":"``hw:numa_cpus.N`` and ``hw:num_mem.N``, refer to the `flavors guide`_."},{"line_number":91,"context_line":""},{"line_number":92,"context_line":"Customizing instance CPU policies"},{"line_number":93,"context_line":"~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~"}],"source_content_type":"text/x-rst","patch_set":12,"id":"dab17558_38834dce","line":90,"range":{"start_line":90,"start_character":54,"end_line":90,"end_character":70},"in_reply_to":"dab17558_8711ac5f","updated":"2016-05-13 09:50:19.000000000","message":"Done","commit_id":"2626a08c524ec267ced6b55cca0c05289ac22510"},{"author":{"_account_id":14962,"name":"ologvinova","email":"ologvinova@mirantis.com","username":"ologvinova"},"change_message_id":"081785ae28507f3a6a2992fb122589990023f088","unresolved":false,"context_lines":[{"line_number":109,"context_line":"instance\u0027s vCPUs. This process is known as pinning. No instance with pinned"},{"line_number":110,"context_line":"CPUs can use the CPUs of another pinned instance, preventing resource"},{"line_number":111,"context_line":"contention between instances. To configure a flavor to use pinned vCPUs, a"},{"line_number":112,"context_line":"dedicated CPU policy should be used. To force this, enable:"},{"line_number":113,"context_line":""},{"line_number":114,"context_line":".. code-block:: console"},{"line_number":115,"context_line":""}],"source_content_type":"text/x-rst","patch_set":12,"id":"dab17558_cae0d92b","line":112,"range":{"start_line":112,"start_character":52,"end_line":112,"end_character":58},"updated":"2016-05-11 11:35:12.000000000","message":"run?","commit_id":"2626a08c524ec267ced6b55cca0c05289ac22510"},{"author":{"_account_id":15334,"name":"Stephen Finucane","display_name":"stephenfin","email":"stephenfin@redhat.com","username":"sfinucan"},"change_message_id":"0c2e44efc3f1d7f40d92f82d1eb3d115935763a2","unresolved":false,"context_lines":[{"line_number":109,"context_line":"instance\u0027s vCPUs. This process is known as pinning. No instance with pinned"},{"line_number":110,"context_line":"CPUs can use the CPUs of another pinned instance, preventing resource"},{"line_number":111,"context_line":"contention between instances. To configure a flavor to use pinned vCPUs, a"},{"line_number":112,"context_line":"dedicated CPU policy should be used. To force this, enable:"},{"line_number":113,"context_line":""},{"line_number":114,"context_line":".. code-block:: console"},{"line_number":115,"context_line":""}],"source_content_type":"text/x-rst","patch_set":12,"id":"dab17558_981b59ca","line":112,"range":{"start_line":112,"start_character":52,"end_line":112,"end_character":58},"in_reply_to":"dab17558_cae0d92b","updated":"2016-05-13 09:50:19.000000000","message":"Done","commit_id":"2626a08c524ec267ced6b55cca0c05289ac22510"},{"author":{"_account_id":14962,"name":"ologvinova","email":"ologvinova@mirantis.com","username":"ologvinova"},"change_message_id":"081785ae28507f3a6a2992fb122589990023f088","unresolved":false,"context_lines":[{"line_number":154,"context_line":"   more information."},{"line_number":155,"context_line":""},{"line_number":156,"context_line":"For more information about the syntax for ``hw:cpu_policy`` and"},{"line_number":157,"context_line":"``hw:cpu_thread_policy``, refer to the `flavors guide`_."},{"line_number":158,"context_line":""},{"line_number":159,"context_line":".. Links"},{"line_number":160,"context_line":".. _`flavors guide`: http://docs.openstack.org/admin-guide/compute-flavors.html"}],"source_content_type":"text/x-rst","patch_set":12,"id":"dab17558_4ad6a928","line":157,"range":{"start_line":157,"start_character":39,"end_line":157,"end_character":55},"updated":"2016-05-11 11:35:12.000000000","message":"`Flavors`_ guide","commit_id":"2626a08c524ec267ced6b55cca0c05289ac22510"},{"author":{"_account_id":15334,"name":"Stephen Finucane","display_name":"stephenfin","email":"stephenfin@redhat.com","username":"sfinucan"},"change_message_id":"0c2e44efc3f1d7f40d92f82d1eb3d115935763a2","unresolved":false,"context_lines":[{"line_number":154,"context_line":"   more information."},{"line_number":155,"context_line":""},{"line_number":156,"context_line":"For more information about the syntax for ``hw:cpu_policy`` and"},{"line_number":157,"context_line":"``hw:cpu_thread_policy``, refer to the `flavors guide`_."},{"line_number":158,"context_line":""},{"line_number":159,"context_line":".. Links"},{"line_number":160,"context_line":".. _`flavors guide`: http://docs.openstack.org/admin-guide/compute-flavors.html"}],"source_content_type":"text/x-rst","patch_set":12,"id":"dab17558_f878a5fa","line":157,"range":{"start_line":157,"start_character":39,"end_line":157,"end_character":55},"in_reply_to":"dab17558_4ad6a928","updated":"2016-05-13 09:50:19.000000000","message":"Done","commit_id":"2626a08c524ec267ced6b55cca0c05289ac22510"},{"author":{"_account_id":14962,"name":"ologvinova","email":"ologvinova@mirantis.com","username":"ologvinova"},"change_message_id":"081785ae28507f3a6a2992fb122589990023f088","unresolved":false,"context_lines":[{"line_number":157,"context_line":"``hw:cpu_thread_policy``, refer to the `flavors guide`_."},{"line_number":158,"context_line":""},{"line_number":159,"context_line":".. Links"},{"line_number":160,"context_line":".. _`flavors guide`: http://docs.openstack.org/admin-guide/compute-flavors.html"}],"source_content_type":"text/x-rst","patch_set":12,"id":"dab17558_2ac0854f","line":160,"range":{"start_line":160,"start_character":4,"end_line":160,"end_character":19},"updated":"2016-05-11 11:35:12.000000000","message":"`Flavors`","commit_id":"2626a08c524ec267ced6b55cca0c05289ac22510"},{"author":{"_account_id":15334,"name":"Stephen Finucane","display_name":"stephenfin","email":"stephenfin@redhat.com","username":"sfinucan"},"change_message_id":"0c2e44efc3f1d7f40d92f82d1eb3d115935763a2","unresolved":false,"context_lines":[{"line_number":157,"context_line":"``hw:cpu_thread_policy``, refer to the `flavors guide`_."},{"line_number":158,"context_line":""},{"line_number":159,"context_line":".. Links"},{"line_number":160,"context_line":".. _`flavors guide`: http://docs.openstack.org/admin-guide/compute-flavors.html"}],"source_content_type":"text/x-rst","patch_set":12,"id":"dab17558_d87b21f3","line":160,"range":{"start_line":160,"start_character":4,"end_line":160,"end_character":19},"in_reply_to":"dab17558_2ac0854f","updated":"2016-05-13 09:50:19.000000000","message":"Done","commit_id":"2626a08c524ec267ced6b55cca0c05289ac22510"},{"author":{"_account_id":6772,"name":"Stephen Gordon","email":"sgordon@redhat.com","username":"sgordon"},"change_message_id":"2ac719b556e4f3e13e7fec19c3e830e4f264d954","unresolved":false,"context_lines":[{"line_number":37,"context_line":".. note::"},{"line_number":38,"context_line":""},{"line_number":39,"context_line":"   The functionality described below is currently only supported by the"},{"line_number":40,"context_line":"   libvirt driver."},{"line_number":41,"context_line":""},{"line_number":42,"context_line":"When running workloads on NUMA hosts, it is important that the vCPUs executing"},{"line_number":43,"context_line":"processes are on the same NUMA node as the memory used by these processes."}],"source_content_type":"text/x-rst","patch_set":13,"id":"dab17558_ea4ae753","line":40,"updated":"2016-05-11 17:04:10.000000000","message":"Do we need to specify libvirt/kvm here (and perhaps even only on some of the archs?) as not all of the libvirt drivers support this do they?","commit_id":"9be283b6e2d37379d464c674543d28228b3b0621"},{"author":{"_account_id":15334,"name":"Stephen Finucane","display_name":"stephenfin","email":"stephenfin@redhat.com","username":"sfinucan"},"change_message_id":"82d695cf76d05cdd90e25e52207681a4a0284f6f","unresolved":false,"context_lines":[{"line_number":37,"context_line":".. note::"},{"line_number":38,"context_line":""},{"line_number":39,"context_line":"   The functionality described below is currently only supported by the"},{"line_number":40,"context_line":"   libvirt driver."},{"line_number":41,"context_line":""},{"line_number":42,"context_line":"When running workloads on NUMA hosts, it is important that the vCPUs executing"},{"line_number":43,"context_line":"processes are on the same NUMA node as the memory used by these processes."}],"source_content_type":"text/x-rst","patch_set":13,"id":"dab17558_f8b9c5ed","line":40,"in_reply_to":"dab17558_ea4ae753","updated":"2016-05-13 12:35:27.000000000","message":"Based on Daniel\u0027s blog [1] it would appear they do?\n\n[1] https://www.berrange.com/posts/2010/02/12/controlling-guest-cpu-numa-affinity-in-libvirt-with-qemu-kvm-xen/","commit_id":"9be283b6e2d37379d464c674543d28228b3b0621"},{"author":{"_account_id":6772,"name":"Stephen Gordon","email":"sgordon@redhat.com","username":"sgordon"},"change_message_id":"2ac719b556e4f3e13e7fec19c3e830e4f264d954","unresolved":false,"context_lines":[{"line_number":57,"context_line":"single NUMA node, then the instance vCPUs will float across a number of host"},{"line_number":58,"context_line":"CPUs and the instance NUMA topology will reflect a multi-node layout. This"},{"line_number":59,"context_line":"allows the operating system of the instance to intelligently schedule the"},{"line_number":60,"context_line":"workloads that it runs."},{"line_number":61,"context_line":""},{"line_number":62,"context_line":"Some workloads have very demanding requirements for memory access latency"},{"line_number":63,"context_line":"and/or bandwidth which exceed the memory bandwidth available from a single"}],"source_content_type":"text/x-rst","patch_set":13,"id":"dab17558_ca456343","line":60,"updated":"2016-05-11 17:04:10.000000000","message":"I think it is worth clarifying what happens with PCIe devices (or their VFs) here as unlike vCPU/vRAM if I request more PCI/SR-IOV resources than are available in a single node the scheduling request will fail if NUMATopologyFilter is enabled (e.g. 2 VFs from different cards that are on different nodes)?","commit_id":"9be283b6e2d37379d464c674543d28228b3b0621"},{"author":{"_account_id":15334,"name":"Stephen Finucane","display_name":"stephenfin","email":"stephenfin@redhat.com","username":"sfinucan"},"change_message_id":"82d695cf76d05cdd90e25e52207681a4a0284f6f","unresolved":false,"context_lines":[{"line_number":57,"context_line":"single NUMA node, then the instance vCPUs will float across a number of host"},{"line_number":58,"context_line":"CPUs and the instance NUMA topology will reflect a multi-node layout. This"},{"line_number":59,"context_line":"allows the operating system of the instance to intelligently schedule the"},{"line_number":60,"context_line":"workloads that it runs."},{"line_number":61,"context_line":""},{"line_number":62,"context_line":"Some workloads have very demanding requirements for memory access latency"},{"line_number":63,"context_line":"and/or bandwidth which exceed the memory bandwidth available from a single"}],"source_content_type":"text/x-rst","patch_set":13,"id":"dab17558_63bfe6fc","line":60,"in_reply_to":"dab17558_ca456343","updated":"2016-05-13 12:35:27.000000000","message":"Very good point. I\u0027ve also included a link to the mailing list discussion [1] where this is brought up, as it thing it provides a lot more info than I can fit here.\n\n[1] http://lists.openstack.org/pipermail/openstack-dev/2016-March/090367.html","commit_id":"9be283b6e2d37379d464c674543d28228b3b0621"},{"author":{"_account_id":6772,"name":"Stephen Gordon","email":"sgordon@redhat.com","username":"sgordon"},"change_message_id":"2ac719b556e4f3e13e7fec19c3e830e4f264d954","unresolved":false,"context_lines":[{"line_number":68,"context_line":""},{"line_number":69,"context_line":".. code-block:: console"},{"line_number":70,"context_line":""},{"line_number":71,"context_line":"   # openstack flavor set m1.large --property hw:numa_nodes\u003d2"},{"line_number":72,"context_line":""},{"line_number":73,"context_line":"The allocation of instances vCPUs and memory from different host NUMA nodes"},{"line_number":74,"context_line":"can be configured. This allows for asymmetric allocation of vCPUs and memory,"}],"source_content_type":"text/x-rst","patch_set":13,"id":"dab17558_2a77cfa0","line":71,"updated":"2016-05-11 17:04:10.000000000","message":"Related to my query above if this works around the PCI/SR-IOV issue that should be highlighted.","commit_id":"9be283b6e2d37379d464c674543d28228b3b0621"},{"author":{"_account_id":15334,"name":"Stephen Finucane","display_name":"stephenfin","email":"stephenfin@redhat.com","username":"sfinucan"},"change_message_id":"82d695cf76d05cdd90e25e52207681a4a0284f6f","unresolved":false,"context_lines":[{"line_number":68,"context_line":""},{"line_number":69,"context_line":".. code-block:: console"},{"line_number":70,"context_line":""},{"line_number":71,"context_line":"   # openstack flavor set m1.large --property hw:numa_nodes\u003d2"},{"line_number":72,"context_line":""},{"line_number":73,"context_line":"The allocation of instances vCPUs and memory from different host NUMA nodes"},{"line_number":74,"context_line":"can be configured. This allows for asymmetric allocation of vCPUs and memory,"}],"source_content_type":"text/x-rst","patch_set":13,"id":"dab17558_c3be92fe","line":71,"in_reply_to":"dab17558_2a77cfa0","updated":"2016-05-13 12:35:27.000000000","message":"Done","commit_id":"9be283b6e2d37379d464c674543d28228b3b0621"},{"author":{"_account_id":6772,"name":"Stephen Gordon","email":"sgordon@redhat.com","username":"sgordon"},"change_message_id":"2ac719b556e4f3e13e7fec19c3e830e4f264d954","unresolved":false,"context_lines":[{"line_number":109,"context_line":"instance\u0027s vCPUs. This process is known as pinning. No instance with pinned"},{"line_number":110,"context_line":"CPUs can use the CPUs of another pinned instance, preventing resource"},{"line_number":111,"context_line":"contention between instances. To configure a flavor to use pinned vCPUs, a"},{"line_number":112,"context_line":"dedicated CPU policy should be used. To force this, enable:"},{"line_number":113,"context_line":""},{"line_number":114,"context_line":".. code-block:: console"},{"line_number":115,"context_line":""}],"source_content_type":"text/x-rst","patch_set":13,"id":"dab17558_0a7acb87","line":112,"updated":"2016-05-11 17:04:10.000000000","message":"Have to highlight the need to use host aggregates in this section, otherwise dedicated and shared/overcommitted guests can and will land on the same host(s) thus effectively removing the usefulness off the dedicated policy.","commit_id":"9be283b6e2d37379d464c674543d28228b3b0621"},{"author":{"_account_id":15334,"name":"Stephen Finucane","display_name":"stephenfin","email":"stephenfin@redhat.com","username":"sfinucan"},"change_message_id":"82d695cf76d05cdd90e25e52207681a4a0284f6f","unresolved":false,"context_lines":[{"line_number":109,"context_line":"instance\u0027s vCPUs. This process is known as pinning. No instance with pinned"},{"line_number":110,"context_line":"CPUs can use the CPUs of another pinned instance, preventing resource"},{"line_number":111,"context_line":"contention between instances. To configure a flavor to use pinned vCPUs, a"},{"line_number":112,"context_line":"dedicated CPU policy should be used. To force this, enable:"},{"line_number":113,"context_line":""},{"line_number":114,"context_line":".. code-block:: console"},{"line_number":115,"context_line":""}],"source_content_type":"text/x-rst","patch_set":13,"id":"dab17558_83680abc","line":112,"in_reply_to":"dab17558_0a7acb87","updated":"2016-05-13 12:35:27.000000000","message":"I already mention this as part of the Flavors guide, but I\u0027ve duplicated it here for posterity.","commit_id":"9be283b6e2d37379d464c674543d28228b3b0621"},{"author":{"_account_id":6772,"name":"Stephen Gordon","email":"sgordon@redhat.com","username":"sgordon"},"change_message_id":"2ac719b556e4f3e13e7fec19c3e830e4f264d954","unresolved":false,"context_lines":[{"line_number":135,"context_line":""},{"line_number":136,"context_line":"   # openstack flavor set m1.large \\"},{"line_number":137,"context_line":"       --property hw:cpu_policy\u003ddedicated \\"},{"line_number":138,"context_line":"       --property hw:cpu_thread_policy\u003dseparate"},{"line_number":139,"context_line":""},{"line_number":140,"context_line":"Finally for workloads where performance is minimally impacted, thread siblings"},{"line_number":141,"context_line":"can be preferred. This is the default, but it can be set explicitly:"}],"source_content_type":"text/x-rst","patch_set":13,"id":"dab17558_6a6d57cb","line":138,"updated":"2016-05-11 17:04:10.000000000","message":"separate isn\u0027t actually a valid value in the final version of this spec?: \n\nhttp://specs.openstack.org/openstack/nova-specs/specs/mitaka/implemented/virt-driver-cpu-thread-pinning.html","commit_id":"9be283b6e2d37379d464c674543d28228b3b0621"},{"author":{"_account_id":15334,"name":"Stephen Finucane","display_name":"stephenfin","email":"stephenfin@redhat.com","username":"sfinucan"},"change_message_id":"82d695cf76d05cdd90e25e52207681a4a0284f6f","unresolved":false,"context_lines":[{"line_number":135,"context_line":""},{"line_number":136,"context_line":"   # openstack flavor set m1.large \\"},{"line_number":137,"context_line":"       --property hw:cpu_policy\u003ddedicated \\"},{"line_number":138,"context_line":"       --property hw:cpu_thread_policy\u003dseparate"},{"line_number":139,"context_line":""},{"line_number":140,"context_line":"Finally for workloads where performance is minimally impacted, thread siblings"},{"line_number":141,"context_line":"can be preferred. This is the default, but it can be set explicitly:"}],"source_content_type":"text/x-rst","patch_set":13,"id":"dab17558_d88a011a","line":138,"in_reply_to":"dab17558_6a6d57cb","updated":"2016-05-13 12:35:27.000000000","message":"Done","commit_id":"9be283b6e2d37379d464c674543d28228b3b0621"},{"author":{"_account_id":10607,"name":"Alexandra Settle","email":"a.settle@outlook.com","username":"asettle"},"change_message_id":"b37c3d83e66245983084970440b1983562472951","unresolved":false,"context_lines":[{"line_number":4,"context_line":"Enabling advanced CPU topologies in guests"},{"line_number":5,"context_line":"\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d"},{"line_number":6,"context_line":""},{"line_number":7,"context_line":"The NUMA topology and CPU pinning features in OpenStack provide a high level"},{"line_number":8,"context_line":"of control over how instances run on host CPUs, and the topology of CPUs"},{"line_number":9,"context_line":"inside the instance. These features can be used to minimize latency and"},{"line_number":10,"context_line":"maximize per-instance performance."},{"line_number":11,"context_line":""}],"source_content_type":"text/x-rst","patch_set":14,"id":"dab17558_928975cd","line":8,"range":{"start_line":7,"start_character":56,"end_line":8,"end_character":29},"updated":"2016-05-13 10:46:48.000000000","message":"provide high level control over instances running* - just a suggestion","commit_id":"adb523c952b7eeab1098bf74102c4c3948c42fb3"},{"author":{"_account_id":15334,"name":"Stephen Finucane","display_name":"stephenfin","email":"stephenfin@redhat.com","username":"sfinucan"},"change_message_id":"a578867c6c058b7379e50290f5113c81f1be8050","unresolved":false,"context_lines":[{"line_number":4,"context_line":"Enabling advanced CPU topologies in guests"},{"line_number":5,"context_line":"\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d"},{"line_number":6,"context_line":""},{"line_number":7,"context_line":"The NUMA topology and CPU pinning features in OpenStack provide a high level"},{"line_number":8,"context_line":"of control over how instances run on host CPUs, and the topology of CPUs"},{"line_number":9,"context_line":"inside the instance. These features can be used to minimize latency and"},{"line_number":10,"context_line":"maximize per-instance performance."},{"line_number":11,"context_line":""}],"source_content_type":"text/x-rst","patch_set":14,"id":"dab17558_5709b72b","line":8,"range":{"start_line":7,"start_character":56,"end_line":8,"end_character":29},"in_reply_to":"dab17558_928975cd","updated":"2016-05-13 12:48:23.000000000","message":"Done","commit_id":"adb523c952b7eeab1098bf74102c4c3948c42fb3"},{"author":{"_account_id":10607,"name":"Alexandra Settle","email":"a.settle@outlook.com","username":"asettle"},"change_message_id":"b37c3d83e66245983084970440b1983562472951","unresolved":false,"context_lines":[{"line_number":20,"context_line":"Non-uniform memory access (NUMA) is a derivative of the SMP design that is"},{"line_number":21,"context_line":"found in many multi-socket systems. In a NUMA system, system memory is divided"},{"line_number":22,"context_line":"into cells or nodes that are associated with particular CPUs. Requests for"},{"line_number":23,"context_line":"memory on other nodes are possible through an interconnect bus; however,"},{"line_number":24,"context_line":"bandwidth across this shared bus is  limited and therefore competition for"},{"line_number":25,"context_line":"this resource can incur performance penalties."},{"line_number":26,"context_line":""}],"source_content_type":"text/x-rst","patch_set":14,"id":"dab17558_523b8deb","line":23,"range":{"start_line":23,"start_character":62,"end_line":23,"end_character":63},"updated":"2016-05-13 10:46:48.000000000","message":"\"Requests for memory on other nodes are possible through an interconnect bus, however, bandwidth across this shared bus...\"","commit_id":"adb523c952b7eeab1098bf74102c4c3948c42fb3"},{"author":{"_account_id":15334,"name":"Stephen Finucane","display_name":"stephenfin","email":"stephenfin@redhat.com","username":"sfinucan"},"change_message_id":"a578867c6c058b7379e50290f5113c81f1be8050","unresolved":false,"context_lines":[{"line_number":20,"context_line":"Non-uniform memory access (NUMA) is a derivative of the SMP design that is"},{"line_number":21,"context_line":"found in many multi-socket systems. In a NUMA system, system memory is divided"},{"line_number":22,"context_line":"into cells or nodes that are associated with particular CPUs. Requests for"},{"line_number":23,"context_line":"memory on other nodes are possible through an interconnect bus; however,"},{"line_number":24,"context_line":"bandwidth across this shared bus is  limited and therefore competition for"},{"line_number":25,"context_line":"this resource can incur performance penalties."},{"line_number":26,"context_line":""}],"source_content_type":"text/x-rst","patch_set":14,"id":"dab17558_d71da767","line":23,"range":{"start_line":23,"start_character":62,"end_line":23,"end_character":63},"in_reply_to":"dab17558_523b8deb","updated":"2016-05-13 12:48:23.000000000","message":"Done","commit_id":"adb523c952b7eeab1098bf74102c4c3948c42fb3"},{"author":{"_account_id":10607,"name":"Alexandra Settle","email":"a.settle@outlook.com","username":"asettle"},"change_message_id":"b37c3d83e66245983084970440b1983562472951","unresolved":false,"context_lines":[{"line_number":21,"context_line":"found in many multi-socket systems. In a NUMA system, system memory is divided"},{"line_number":22,"context_line":"into cells or nodes that are associated with particular CPUs. Requests for"},{"line_number":23,"context_line":"memory on other nodes are possible through an interconnect bus; however,"},{"line_number":24,"context_line":"bandwidth across this shared bus is  limited and therefore competition for"},{"line_number":25,"context_line":"this resource can incur performance penalties."},{"line_number":26,"context_line":""},{"line_number":27,"context_line":"Simultaneous Multi-Threading (SMT), known as as Hyperthreading on Intel"}],"source_content_type":"text/x-rst","patch_set":14,"id":"dab17558_1218e583","line":24,"range":{"start_line":24,"start_character":36,"end_line":24,"end_character":37},"updated":"2016-05-13 10:46:48.000000000","message":"Remove the extra space here","commit_id":"adb523c952b7eeab1098bf74102c4c3948c42fb3"},{"author":{"_account_id":15334,"name":"Stephen Finucane","display_name":"stephenfin","email":"stephenfin@redhat.com","username":"sfinucan"},"change_message_id":"a578867c6c058b7379e50290f5113c81f1be8050","unresolved":false,"context_lines":[{"line_number":21,"context_line":"found in many multi-socket systems. In a NUMA system, system memory is divided"},{"line_number":22,"context_line":"into cells or nodes that are associated with particular CPUs. Requests for"},{"line_number":23,"context_line":"memory on other nodes are possible through an interconnect bus; however,"},{"line_number":24,"context_line":"bandwidth across this shared bus is  limited and therefore competition for"},{"line_number":25,"context_line":"this resource can incur performance penalties."},{"line_number":26,"context_line":""},{"line_number":27,"context_line":"Simultaneous Multi-Threading (SMT), known as as Hyperthreading on Intel"}],"source_content_type":"text/x-rst","patch_set":14,"id":"dab17558_17260fba","line":24,"range":{"start_line":24,"start_character":36,"end_line":24,"end_character":37},"in_reply_to":"dab17558_1218e583","updated":"2016-05-13 12:48:23.000000000","message":"Done","commit_id":"adb523c952b7eeab1098bf74102c4c3948c42fb3"},{"author":{"_account_id":10607,"name":"Alexandra Settle","email":"a.settle@outlook.com","username":"asettle"},"change_message_id":"b37c3d83e66245983084970440b1983562472951","unresolved":false,"context_lines":[{"line_number":28,"context_line":"platforms, is an design that is complementary to SMP. Whereas CPUs in SMP"},{"line_number":29,"context_line":"systems share a bus and some memory, CPUs in SMT systems share many more"},{"line_number":30,"context_line":"components. CPUs that share components are known as thread siblings. All CPUs"},{"line_number":31,"context_line":"appear as usable CPUs on the system and can execute workloads in parallel;"},{"line_number":32,"context_line":"however, as with NUMA, threads compete for shared resources."},{"line_number":33,"context_line":""},{"line_number":34,"context_line":"Customizing instance NUMA topologies"}],"source_content_type":"text/x-rst","patch_set":14,"id":"dab17558_350cf33a","line":31,"range":{"start_line":31,"start_character":73,"end_line":31,"end_character":74},"updated":"2016-05-13 10:46:48.000000000","message":"A comma is fine here too","commit_id":"adb523c952b7eeab1098bf74102c4c3948c42fb3"},{"author":{"_account_id":15334,"name":"Stephen Finucane","display_name":"stephenfin","email":"stephenfin@redhat.com","username":"sfinucan"},"change_message_id":"a578867c6c058b7379e50290f5113c81f1be8050","unresolved":false,"context_lines":[{"line_number":28,"context_line":"platforms, is an design that is complementary to SMP. Whereas CPUs in SMP"},{"line_number":29,"context_line":"systems share a bus and some memory, CPUs in SMT systems share many more"},{"line_number":30,"context_line":"components. CPUs that share components are known as thread siblings. All CPUs"},{"line_number":31,"context_line":"appear as usable CPUs on the system and can execute workloads in parallel;"},{"line_number":32,"context_line":"however, as with NUMA, threads compete for shared resources."},{"line_number":33,"context_line":""},{"line_number":34,"context_line":"Customizing instance NUMA topologies"}],"source_content_type":"text/x-rst","patch_set":14,"id":"dab17558_d73487df","line":31,"range":{"start_line":31,"start_character":73,"end_line":31,"end_character":74},"in_reply_to":"dab17558_350cf33a","updated":"2016-05-13 12:48:23.000000000","message":"Done","commit_id":"adb523c952b7eeab1098bf74102c4c3948c42fb3"},{"author":{"_account_id":10607,"name":"Alexandra Settle","email":"a.settle@outlook.com","username":"asettle"},"change_message_id":"b37c3d83e66245983084970440b1983562472951","unresolved":false,"context_lines":[{"line_number":43,"context_line":"processes are on the same NUMA node as the memory used by these processes."},{"line_number":44,"context_line":"This ensures all memory accesses are local to the node and thus do not consume"},{"line_number":45,"context_line":"the limited cross-node memory bandwidth, adding latency to memory accesses."},{"line_number":46,"context_line":"Similarly, hugepages are assigned from memory and benefit from the same"},{"line_number":47,"context_line":"performance improvements as memory allocated using standard pages, thus, they"},{"line_number":48,"context_line":"also should be local. Finally, PCI devices are directly associated with"},{"line_number":49,"context_line":"specific NUMA nodes for the purposes of DMA. When using PCI device assignment,"}],"source_content_type":"text/x-rst","patch_set":14,"id":"dab17558_f54f0b68","line":46,"range":{"start_line":46,"start_character":11,"end_line":46,"end_character":20},"updated":"2016-05-13 10:46:48.000000000","message":"I believe this is meant to be two words? I can\u0027t find any consistency surrounding it though.","commit_id":"adb523c952b7eeab1098bf74102c4c3948c42fb3"},{"author":{"_account_id":15334,"name":"Stephen Finucane","display_name":"stephenfin","email":"stephenfin@redhat.com","username":"sfinucan"},"change_message_id":"a578867c6c058b7379e50290f5113c81f1be8050","unresolved":false,"context_lines":[{"line_number":43,"context_line":"processes are on the same NUMA node as the memory used by these processes."},{"line_number":44,"context_line":"This ensures all memory accesses are local to the node and thus do not consume"},{"line_number":45,"context_line":"the limited cross-node memory bandwidth, adding latency to memory accesses."},{"line_number":46,"context_line":"Similarly, hugepages are assigned from memory and benefit from the same"},{"line_number":47,"context_line":"performance improvements as memory allocated using standard pages, thus, they"},{"line_number":48,"context_line":"also should be local. Finally, PCI devices are directly associated with"},{"line_number":49,"context_line":"specific NUMA nodes for the purposes of DMA. When using PCI device assignment,"}],"source_content_type":"text/x-rst","patch_set":14,"id":"dab17558_d74be757","line":46,"range":{"start_line":46,"start_character":11,"end_line":46,"end_character":20},"in_reply_to":"dab17558_f54f0b68","updated":"2016-05-13 12:48:23.000000000","message":"The flavors guide uses \"large pages\" so I used this","commit_id":"adb523c952b7eeab1098bf74102c4c3948c42fb3"},{"author":{"_account_id":10607,"name":"Alexandra Settle","email":"a.settle@outlook.com","username":"asettle"},"change_message_id":"b37c3d83e66245983084970440b1983562472951","unresolved":false,"context_lines":[{"line_number":46,"context_line":"Similarly, hugepages are assigned from memory and benefit from the same"},{"line_number":47,"context_line":"performance improvements as memory allocated using standard pages, thus, they"},{"line_number":48,"context_line":"also should be local. Finally, PCI devices are directly associated with"},{"line_number":49,"context_line":"specific NUMA nodes for the purposes of DMA. When using PCI device assignment,"},{"line_number":50,"context_line":"it is also desirable that the instance be placed on the same NUMA node as any"},{"line_number":51,"context_line":"PCI device that is assigned to it."},{"line_number":52,"context_line":""},{"line_number":53,"context_line":"By default, if the RAM and vCPUs associated with a flavor can fit onto a"},{"line_number":54,"context_line":"single NUMA node, then the vCPUs of that instance will only float across the"}],"source_content_type":"text/x-rst","patch_set":14,"id":"dab17558_753cbbbf","line":51,"range":{"start_line":49,"start_character":45,"end_line":51,"end_character":34},"updated":"2016-05-13 10:46:48.000000000","message":"Fix the passive voice here","commit_id":"adb523c952b7eeab1098bf74102c4c3948c42fb3"},{"author":{"_account_id":15334,"name":"Stephen Finucane","display_name":"stephenfin","email":"stephenfin@redhat.com","username":"sfinucan"},"change_message_id":"a578867c6c058b7379e50290f5113c81f1be8050","unresolved":false,"context_lines":[{"line_number":46,"context_line":"Similarly, hugepages are assigned from memory and benefit from the same"},{"line_number":47,"context_line":"performance improvements as memory allocated using standard pages, thus, they"},{"line_number":48,"context_line":"also should be local. Finally, PCI devices are directly associated with"},{"line_number":49,"context_line":"specific NUMA nodes for the purposes of DMA. When using PCI device assignment,"},{"line_number":50,"context_line":"it is also desirable that the instance be placed on the same NUMA node as any"},{"line_number":51,"context_line":"PCI device that is assigned to it."},{"line_number":52,"context_line":""},{"line_number":53,"context_line":"By default, if the RAM and vCPUs associated with a flavor can fit onto a"},{"line_number":54,"context_line":"single NUMA node, then the vCPUs of that instance will only float across the"}],"source_content_type":"text/x-rst","patch_set":14,"id":"dab17558_57f3770a","line":51,"range":{"start_line":49,"start_character":45,"end_line":51,"end_character":34},"in_reply_to":"dab17558_753cbbbf","updated":"2016-05-13 12:48:23.000000000","message":"Think I\u0027ve done this. Feel free to reword as you see fit.","commit_id":"adb523c952b7eeab1098bf74102c4c3948c42fb3"},{"author":{"_account_id":10607,"name":"Alexandra Settle","email":"a.settle@outlook.com","username":"asettle"},"change_message_id":"b37c3d83e66245983084970440b1983562472951","unresolved":false,"context_lines":[{"line_number":51,"context_line":"PCI device that is assigned to it."},{"line_number":52,"context_line":""},{"line_number":53,"context_line":"By default, if the RAM and vCPUs associated with a flavor can fit onto a"},{"line_number":54,"context_line":"single NUMA node, then the vCPUs of that instance will only float across the"},{"line_number":55,"context_line":"host CPUs on a given node and the instance NUMA topology will reflect a single"},{"line_number":56,"context_line":"node layout. If the RAM or vCPUs associated with a flavor are larger than any"},{"line_number":57,"context_line":"single NUMA node, then the instance vCPUs will float across a number of host"}],"source_content_type":"text/x-rst","patch_set":14,"id":"dab17558_f576eba1","line":54,"range":{"start_line":54,"start_character":50,"end_line":54,"end_character":59},"updated":"2016-05-13 10:46:48.000000000","message":"only floats*","commit_id":"adb523c952b7eeab1098bf74102c4c3948c42fb3"},{"author":{"_account_id":15334,"name":"Stephen Finucane","display_name":"stephenfin","email":"stephenfin@redhat.com","username":"sfinucan"},"change_message_id":"a578867c6c058b7379e50290f5113c81f1be8050","unresolved":false,"context_lines":[{"line_number":51,"context_line":"PCI device that is assigned to it."},{"line_number":52,"context_line":""},{"line_number":53,"context_line":"By default, if the RAM and vCPUs associated with a flavor can fit onto a"},{"line_number":54,"context_line":"single NUMA node, then the vCPUs of that instance will only float across the"},{"line_number":55,"context_line":"host CPUs on a given node and the instance NUMA topology will reflect a single"},{"line_number":56,"context_line":"node layout. If the RAM or vCPUs associated with a flavor are larger than any"},{"line_number":57,"context_line":"single NUMA node, then the instance vCPUs will float across a number of host"}],"source_content_type":"text/x-rst","patch_set":14,"id":"dab17558_77f4bbce","line":54,"range":{"start_line":54,"start_character":50,"end_line":54,"end_character":59},"in_reply_to":"dab17558_f576eba1","updated":"2016-05-13 12:48:23.000000000","message":"Done","commit_id":"adb523c952b7eeab1098bf74102c4c3948c42fb3"},{"author":{"_account_id":10607,"name":"Alexandra Settle","email":"a.settle@outlook.com","username":"asettle"},"change_message_id":"b37c3d83e66245983084970440b1983562472951","unresolved":false,"context_lines":[{"line_number":54,"context_line":"single NUMA node, then the vCPUs of that instance will only float across the"},{"line_number":55,"context_line":"host CPUs on a given node and the instance NUMA topology will reflect a single"},{"line_number":56,"context_line":"node layout. If the RAM or vCPUs associated with a flavor are larger than any"},{"line_number":57,"context_line":"single NUMA node, then the instance vCPUs will float across a number of host"},{"line_number":58,"context_line":"CPUs and the instance NUMA topology will reflect a multi-node layout. This"},{"line_number":59,"context_line":"allows the operating system of the instance to intelligently schedule the"},{"line_number":60,"context_line":"workloads that it runs."}],"source_content_type":"text/x-rst","patch_set":14,"id":"dab17558_357193a9","line":57,"range":{"start_line":57,"start_character":42,"end_line":57,"end_character":52},"updated":"2016-05-13 10:46:48.000000000","message":"ditto","commit_id":"adb523c952b7eeab1098bf74102c4c3948c42fb3"},{"author":{"_account_id":15334,"name":"Stephen Finucane","display_name":"stephenfin","email":"stephenfin@redhat.com","username":"sfinucan"},"change_message_id":"a578867c6c058b7379e50290f5113c81f1be8050","unresolved":false,"context_lines":[{"line_number":54,"context_line":"single NUMA node, then the vCPUs of that instance will only float across the"},{"line_number":55,"context_line":"host CPUs on a given node and the instance NUMA topology will reflect a single"},{"line_number":56,"context_line":"node layout. If the RAM or vCPUs associated with a flavor are larger than any"},{"line_number":57,"context_line":"single NUMA node, then the instance vCPUs will float across a number of host"},{"line_number":58,"context_line":"CPUs and the instance NUMA topology will reflect a multi-node layout. This"},{"line_number":59,"context_line":"allows the operating system of the instance to intelligently schedule the"},{"line_number":60,"context_line":"workloads that it runs."}],"source_content_type":"text/x-rst","patch_set":14,"id":"dab17558_57b417f1","line":57,"range":{"start_line":57,"start_character":42,"end_line":57,"end_character":52},"in_reply_to":"dab17558_357193a9","updated":"2016-05-13 12:48:23.000000000","message":"Done","commit_id":"adb523c952b7eeab1098bf74102c4c3948c42fb3"},{"author":{"_account_id":10607,"name":"Alexandra Settle","email":"a.settle@outlook.com","username":"asettle"},"change_message_id":"b37c3d83e66245983084970440b1983562472951","unresolved":false,"context_lines":[{"line_number":60,"context_line":"workloads that it runs."},{"line_number":61,"context_line":""},{"line_number":62,"context_line":"Some workloads have very demanding requirements for memory access latency"},{"line_number":63,"context_line":"and/or bandwidth which exceed the memory bandwidth available from a single"},{"line_number":64,"context_line":"NUMA node. For such workloads, it is beneficial to spread the instance across"},{"line_number":65,"context_line":"multiple host NUMA nodes, even if the instance\u0027s RAM/vCPUs could theoretically"},{"line_number":66,"context_line":"fit on a single NUMA node. To force an instance\u0027s VCPUs to spread across two"}],"source_content_type":"text/x-rst","patch_set":14,"id":"dab17558_b55d8317","line":63,"range":{"start_line":63,"start_character":0,"end_line":63,"end_character":6},"updated":"2016-05-13 10:46:48.000000000","message":"Remove the slash and pick one :)","commit_id":"adb523c952b7eeab1098bf74102c4c3948c42fb3"},{"author":{"_account_id":15334,"name":"Stephen Finucane","display_name":"stephenfin","email":"stephenfin@redhat.com","username":"sfinucan"},"change_message_id":"a578867c6c058b7379e50290f5113c81f1be8050","unresolved":false,"context_lines":[{"line_number":60,"context_line":"workloads that it runs."},{"line_number":61,"context_line":""},{"line_number":62,"context_line":"Some workloads have very demanding requirements for memory access latency"},{"line_number":63,"context_line":"and/or bandwidth which exceed the memory bandwidth available from a single"},{"line_number":64,"context_line":"NUMA node. For such workloads, it is beneficial to spread the instance across"},{"line_number":65,"context_line":"multiple host NUMA nodes, even if the instance\u0027s RAM/vCPUs could theoretically"},{"line_number":66,"context_line":"fit on a single NUMA node. To force an instance\u0027s VCPUs to spread across two"}],"source_content_type":"text/x-rst","patch_set":14,"id":"dab17558_f7bc2bdd","line":63,"range":{"start_line":63,"start_character":0,"end_line":63,"end_character":6},"in_reply_to":"dab17558_b55d8317","updated":"2016-05-13 12:48:23.000000000","message":"Done","commit_id":"adb523c952b7eeab1098bf74102c4c3948c42fb3"},{"author":{"_account_id":10607,"name":"Alexandra Settle","email":"a.settle@outlook.com","username":"asettle"},"change_message_id":"b37c3d83e66245983084970440b1983562472951","unresolved":false,"context_lines":[{"line_number":70,"context_line":""},{"line_number":71,"context_line":"   # openstack flavor set m1.large --property hw:numa_nodes\u003d2"},{"line_number":72,"context_line":""},{"line_number":73,"context_line":"The allocation of instances vCPUs and memory from different host NUMA nodes"},{"line_number":74,"context_line":"can be configured. This allows for asymmetric allocation of vCPUs and memory,"},{"line_number":75,"context_line":"which can be important for some workloads. To spread the six vCPUs and 6GB of"},{"line_number":76,"context_line":"memory of an instance across two NUMA nodes and create an asymmetric 1:2 vCPU"},{"line_number":77,"context_line":"and memory mapping between the two nodes, run:"}],"source_content_type":"text/x-rst","patch_set":14,"id":"dab17558_d5b92730","line":74,"range":{"start_line":73,"start_character":0,"end_line":74,"end_character":18},"updated":"2016-05-13 10:46:48.000000000","message":"Configure the allocation...","commit_id":"adb523c952b7eeab1098bf74102c4c3948c42fb3"},{"author":{"_account_id":15334,"name":"Stephen Finucane","display_name":"stephenfin","email":"stephenfin@redhat.com","username":"sfinucan"},"change_message_id":"a578867c6c058b7379e50290f5113c81f1be8050","unresolved":false,"context_lines":[{"line_number":70,"context_line":""},{"line_number":71,"context_line":"   # openstack flavor set m1.large --property hw:numa_nodes\u003d2"},{"line_number":72,"context_line":""},{"line_number":73,"context_line":"The allocation of instances vCPUs and memory from different host NUMA nodes"},{"line_number":74,"context_line":"can be configured. This allows for asymmetric allocation of vCPUs and memory,"},{"line_number":75,"context_line":"which can be important for some workloads. To spread the six vCPUs and 6GB of"},{"line_number":76,"context_line":"memory of an instance across two NUMA nodes and create an asymmetric 1:2 vCPU"},{"line_number":77,"context_line":"and memory mapping between the two nodes, run:"}],"source_content_type":"text/x-rst","patch_set":14,"id":"dab17558_17b60fdc","line":74,"range":{"start_line":73,"start_character":0,"end_line":74,"end_character":18},"in_reply_to":"dab17558_d5b92730","updated":"2016-05-13 12:48:23.000000000","message":"I get what you\u0027re trying to avoid, but this isn\u0027t correct. This step is optional idea, rather than something that must be done. In many cases, this won\u0027t be done. Open to suggestions here.","commit_id":"adb523c952b7eeab1098bf74102c4c3948c42fb3"},{"author":{"_account_id":10607,"name":"Alexandra Settle","email":"a.settle@outlook.com","username":"asettle"},"change_message_id":"b37c3d83e66245983084970440b1983562472951","unresolved":false,"context_lines":[{"line_number":98,"context_line":"   libvirt driver."},{"line_number":99,"context_line":""},{"line_number":100,"context_line":"By default, instance vCPU processes are not assigned to any particular host"},{"line_number":101,"context_line":"CPU; instead, they float across host CPUs like any other process. This allows"},{"line_number":102,"context_line":"for features like overcommitting of CPUs. In heavily contended systems, this"},{"line_number":103,"context_line":"provides optimal system performance at the expense of performance and latency"},{"line_number":104,"context_line":"for individual instances."}],"source_content_type":"text/x-rst","patch_set":14,"id":"dab17558_75d83b80","line":101,"range":{"start_line":101,"start_character":3,"end_line":101,"end_character":4},"updated":"2016-05-13 10:46:48.000000000","message":"Please refrain from using semi-colons :)","commit_id":"adb523c952b7eeab1098bf74102c4c3948c42fb3"},{"author":{"_account_id":15334,"name":"Stephen Finucane","display_name":"stephenfin","email":"stephenfin@redhat.com","username":"sfinucan"},"change_message_id":"a578867c6c058b7379e50290f5113c81f1be8050","unresolved":false,"context_lines":[{"line_number":98,"context_line":"   libvirt driver."},{"line_number":99,"context_line":""},{"line_number":100,"context_line":"By default, instance vCPU processes are not assigned to any particular host"},{"line_number":101,"context_line":"CPU; instead, they float across host CPUs like any other process. This allows"},{"line_number":102,"context_line":"for features like overcommitting of CPUs. In heavily contended systems, this"},{"line_number":103,"context_line":"provides optimal system performance at the expense of performance and latency"},{"line_number":104,"context_line":"for individual instances."}],"source_content_type":"text/x-rst","patch_set":14,"id":"dab17558_37b353eb","line":101,"range":{"start_line":101,"start_character":3,"end_line":101,"end_character":4},"in_reply_to":"dab17558_75d83b80","updated":"2016-05-13 12:48:23.000000000","message":"Done","commit_id":"adb523c952b7eeab1098bf74102c4c3948c42fb3"},{"author":{"_account_id":10607,"name":"Alexandra Settle","email":"a.settle@outlook.com","username":"asettle"},"change_message_id":"b37c3d83e66245983084970440b1983562472951","unresolved":false,"context_lines":[{"line_number":103,"context_line":"provides optimal system performance at the expense of performance and latency"},{"line_number":104,"context_line":"for individual instances."},{"line_number":105,"context_line":""},{"line_number":106,"context_line":"Some workloads must exhibit real-time or near real-time behavior, which would"},{"line_number":107,"context_line":"not be possible with the latency introduced by the default CPU policy. For"},{"line_number":108,"context_line":"such workloads, it is beneficial to control which host CPUs are bound to an"},{"line_number":109,"context_line":"instance\u0027s vCPUs. This process is known as pinning. No instance with pinned"}],"source_content_type":"text/x-rst","patch_set":14,"id":"dab17558_f5d26b5a","line":106,"range":{"start_line":106,"start_character":15,"end_line":106,"end_character":20},"updated":"2016-05-13 10:46:48.000000000","message":"remove","commit_id":"adb523c952b7eeab1098bf74102c4c3948c42fb3"},{"author":{"_account_id":15334,"name":"Stephen Finucane","display_name":"stephenfin","email":"stephenfin@redhat.com","username":"sfinucan"},"change_message_id":"a578867c6c058b7379e50290f5113c81f1be8050","unresolved":false,"context_lines":[{"line_number":103,"context_line":"provides optimal system performance at the expense of performance and latency"},{"line_number":104,"context_line":"for individual instances."},{"line_number":105,"context_line":""},{"line_number":106,"context_line":"Some workloads must exhibit real-time or near real-time behavior, which would"},{"line_number":107,"context_line":"not be possible with the latency introduced by the default CPU policy. For"},{"line_number":108,"context_line":"such workloads, it is beneficial to control which host CPUs are bound to an"},{"line_number":109,"context_line":"instance\u0027s vCPUs. This process is known as pinning. No instance with pinned"}],"source_content_type":"text/x-rst","patch_set":14,"id":"dab17558_f7df0b0a","line":106,"range":{"start_line":106,"start_character":15,"end_line":106,"end_character":20},"in_reply_to":"dab17558_f5d26b5a","updated":"2016-05-13 12:48:23.000000000","message":"Done","commit_id":"adb523c952b7eeab1098bf74102c4c3948c42fb3"},{"author":{"_account_id":10607,"name":"Alexandra Settle","email":"a.settle@outlook.com","username":"asettle"},"change_message_id":"b37c3d83e66245983084970440b1983562472951","unresolved":false,"context_lines":[{"line_number":103,"context_line":"provides optimal system performance at the expense of performance and latency"},{"line_number":104,"context_line":"for individual instances."},{"line_number":105,"context_line":""},{"line_number":106,"context_line":"Some workloads must exhibit real-time or near real-time behavior, which would"},{"line_number":107,"context_line":"not be possible with the latency introduced by the default CPU policy. For"},{"line_number":108,"context_line":"such workloads, it is beneficial to control which host CPUs are bound to an"},{"line_number":109,"context_line":"instance\u0027s vCPUs. This process is known as pinning. No instance with pinned"},{"line_number":110,"context_line":"CPUs can use the CPUs of another pinned instance, preventing resource"}],"source_content_type":"text/x-rst","patch_set":14,"id":"dab17558_75bf9bfc","line":107,"range":{"start_line":106,"start_character":66,"end_line":107,"end_character":6},"updated":"2016-05-13 10:46:48.000000000","message":"which is not possible*","commit_id":"adb523c952b7eeab1098bf74102c4c3948c42fb3"},{"author":{"_account_id":15334,"name":"Stephen Finucane","display_name":"stephenfin","email":"stephenfin@redhat.com","username":"sfinucan"},"change_message_id":"a578867c6c058b7379e50290f5113c81f1be8050","unresolved":false,"context_lines":[{"line_number":103,"context_line":"provides optimal system performance at the expense of performance and latency"},{"line_number":104,"context_line":"for individual instances."},{"line_number":105,"context_line":""},{"line_number":106,"context_line":"Some workloads must exhibit real-time or near real-time behavior, which would"},{"line_number":107,"context_line":"not be possible with the latency introduced by the default CPU policy. For"},{"line_number":108,"context_line":"such workloads, it is beneficial to control which host CPUs are bound to an"},{"line_number":109,"context_line":"instance\u0027s vCPUs. This process is known as pinning. No instance with pinned"},{"line_number":110,"context_line":"CPUs can use the CPUs of another pinned instance, preventing resource"}],"source_content_type":"text/x-rst","patch_set":14,"id":"dab17558_17dd6f14","line":107,"range":{"start_line":106,"start_character":66,"end_line":107,"end_character":6},"in_reply_to":"dab17558_75bf9bfc","updated":"2016-05-13 12:48:23.000000000","message":"Done","commit_id":"adb523c952b7eeab1098bf74102c4c3948c42fb3"},{"author":{"_account_id":10607,"name":"Alexandra Settle","email":"a.settle@outlook.com","username":"asettle"},"change_message_id":"b37c3d83e66245983084970440b1983562472951","unresolved":false,"context_lines":[{"line_number":109,"context_line":"instance\u0027s vCPUs. This process is known as pinning. No instance with pinned"},{"line_number":110,"context_line":"CPUs can use the CPUs of another pinned instance, preventing resource"},{"line_number":111,"context_line":"contention between instances. To configure a flavor to use pinned vCPUs, a"},{"line_number":112,"context_line":"dedicated CPU policy should be used. To force this, enable:"},{"line_number":113,"context_line":""},{"line_number":114,"context_line":".. code-block:: console"},{"line_number":115,"context_line":""}],"source_content_type":"text/x-rst","patch_set":14,"id":"dab17558_5541b70a","line":112,"range":{"start_line":112,"start_character":0,"end_line":112,"end_character":36},"updated":"2016-05-13 10:46:48.000000000","message":"use a dedicated CPU policy*","commit_id":"adb523c952b7eeab1098bf74102c4c3948c42fb3"},{"author":{"_account_id":15334,"name":"Stephen Finucane","display_name":"stephenfin","email":"stephenfin@redhat.com","username":"sfinucan"},"change_message_id":"a578867c6c058b7379e50290f5113c81f1be8050","unresolved":false,"context_lines":[{"line_number":109,"context_line":"instance\u0027s vCPUs. This process is known as pinning. No instance with pinned"},{"line_number":110,"context_line":"CPUs can use the CPUs of another pinned instance, preventing resource"},{"line_number":111,"context_line":"contention between instances. To configure a flavor to use pinned vCPUs, a"},{"line_number":112,"context_line":"dedicated CPU policy should be used. To force this, enable:"},{"line_number":113,"context_line":""},{"line_number":114,"context_line":".. code-block:: console"},{"line_number":115,"context_line":""}],"source_content_type":"text/x-rst","patch_set":14,"id":"dab17558_f72d4bfe","line":112,"range":{"start_line":112,"start_character":0,"end_line":112,"end_character":36},"in_reply_to":"dab17558_5541b70a","updated":"2016-05-13 12:48:23.000000000","message":"Done","commit_id":"adb523c952b7eeab1098bf74102c4c3948c42fb3"},{"author":{"_account_id":10607,"name":"Alexandra Settle","email":"a.settle@outlook.com","username":"asettle"},"change_message_id":"b37c3d83e66245983084970440b1983562472951","unresolved":false,"context_lines":[{"line_number":116,"context_line":"   # openstack flavor set m1.large --property hw:cpu_policy\u003ddedicated"},{"line_number":117,"context_line":""},{"line_number":118,"context_line":"When running workloads on SMT hosts, it is important to be aware of the impact"},{"line_number":119,"context_line":"that thread siblings can have. As stated above, thread siblings share a number"},{"line_number":120,"context_line":"of components and contention on these components can impact performance. To"},{"line_number":121,"context_line":"configure how threads should be used, a CPU thread policy should be specified."},{"line_number":122,"context_line":"For workloads where sharing benefits performance, thread siblings should be"}],"source_content_type":"text/x-rst","patch_set":14,"id":"dab17558_d555a7c6","line":119,"range":{"start_line":119,"start_character":31,"end_line":119,"end_character":49},"updated":"2016-05-13 10:46:48.000000000","message":"Remove \"Thread siblibgs...*","commit_id":"adb523c952b7eeab1098bf74102c4c3948c42fb3"},{"author":{"_account_id":15334,"name":"Stephen Finucane","display_name":"stephenfin","email":"stephenfin@redhat.com","username":"sfinucan"},"change_message_id":"a578867c6c058b7379e50290f5113c81f1be8050","unresolved":false,"context_lines":[{"line_number":116,"context_line":"   # openstack flavor set m1.large --property hw:cpu_policy\u003ddedicated"},{"line_number":117,"context_line":""},{"line_number":118,"context_line":"When running workloads on SMT hosts, it is important to be aware of the impact"},{"line_number":119,"context_line":"that thread siblings can have. As stated above, thread siblings share a number"},{"line_number":120,"context_line":"of components and contention on these components can impact performance. To"},{"line_number":121,"context_line":"configure how threads should be used, a CPU thread policy should be specified."},{"line_number":122,"context_line":"For workloads where sharing benefits performance, thread siblings should be"}],"source_content_type":"text/x-rst","patch_set":14,"id":"dab17558_578377cc","line":119,"range":{"start_line":119,"start_character":31,"end_line":119,"end_character":49},"in_reply_to":"dab17558_d555a7c6","updated":"2016-05-13 12:48:23.000000000","message":"Done","commit_id":"adb523c952b7eeab1098bf74102c4c3948c42fb3"},{"author":{"_account_id":10607,"name":"Alexandra Settle","email":"a.settle@outlook.com","username":"asettle"},"change_message_id":"b37c3d83e66245983084970440b1983562472951","unresolved":false,"context_lines":[{"line_number":118,"context_line":"When running workloads on SMT hosts, it is important to be aware of the impact"},{"line_number":119,"context_line":"that thread siblings can have. As stated above, thread siblings share a number"},{"line_number":120,"context_line":"of components and contention on these components can impact performance. To"},{"line_number":121,"context_line":"configure how threads should be used, a CPU thread policy should be specified."},{"line_number":122,"context_line":"For workloads where sharing benefits performance, thread siblings should be"},{"line_number":123,"context_line":"used. To force this, run:"},{"line_number":124,"context_line":""}],"source_content_type":"text/x-rst","patch_set":14,"id":"dab17558_b56743b6","line":121,"range":{"start_line":121,"start_character":11,"end_line":121,"end_character":36},"updated":"2016-05-13 10:46:48.000000000","message":"how to use threads*","commit_id":"adb523c952b7eeab1098bf74102c4c3948c42fb3"},{"author":{"_account_id":15334,"name":"Stephen Finucane","display_name":"stephenfin","email":"stephenfin@redhat.com","username":"sfinucan"},"change_message_id":"a578867c6c058b7379e50290f5113c81f1be8050","unresolved":false,"context_lines":[{"line_number":118,"context_line":"When running workloads on SMT hosts, it is important to be aware of the impact"},{"line_number":119,"context_line":"that thread siblings can have. As stated above, thread siblings share a number"},{"line_number":120,"context_line":"of components and contention on these components can impact performance. To"},{"line_number":121,"context_line":"configure how threads should be used, a CPU thread policy should be specified."},{"line_number":122,"context_line":"For workloads where sharing benefits performance, thread siblings should be"},{"line_number":123,"context_line":"used. To force this, run:"},{"line_number":124,"context_line":""}],"source_content_type":"text/x-rst","patch_set":14,"id":"dab17558_f77b8bf2","line":121,"range":{"start_line":121,"start_character":11,"end_line":121,"end_character":36},"in_reply_to":"dab17558_b56743b6","updated":"2016-05-13 12:48:23.000000000","message":"Done","commit_id":"adb523c952b7eeab1098bf74102c4c3948c42fb3"},{"author":{"_account_id":10607,"name":"Alexandra Settle","email":"a.settle@outlook.com","username":"asettle"},"change_message_id":"b37c3d83e66245983084970440b1983562472951","unresolved":false,"context_lines":[{"line_number":120,"context_line":"of components and contention on these components can impact performance. To"},{"line_number":121,"context_line":"configure how threads should be used, a CPU thread policy should be specified."},{"line_number":122,"context_line":"For workloads where sharing benefits performance, thread siblings should be"},{"line_number":123,"context_line":"used. To force this, run:"},{"line_number":124,"context_line":""},{"line_number":125,"context_line":".. code-block:: console"},{"line_number":126,"context_line":""}],"source_content_type":"text/x-rst","patch_set":14,"id":"dab17558_3512330a","line":123,"range":{"start_line":123,"start_character":0,"end_line":123,"end_character":5},"updated":"2016-05-13 10:46:48.000000000","message":"same as above*","commit_id":"adb523c952b7eeab1098bf74102c4c3948c42fb3"},{"author":{"_account_id":15334,"name":"Stephen Finucane","display_name":"stephenfin","email":"stephenfin@redhat.com","username":"sfinucan"},"change_message_id":"a578867c6c058b7379e50290f5113c81f1be8050","unresolved":false,"context_lines":[{"line_number":120,"context_line":"of components and contention on these components can impact performance. To"},{"line_number":121,"context_line":"configure how threads should be used, a CPU thread policy should be specified."},{"line_number":122,"context_line":"For workloads where sharing benefits performance, thread siblings should be"},{"line_number":123,"context_line":"used. To force this, run:"},{"line_number":124,"context_line":""},{"line_number":125,"context_line":".. code-block:: console"},{"line_number":126,"context_line":""}],"source_content_type":"text/x-rst","patch_set":14,"id":"dab17558_3786b3db","line":123,"range":{"start_line":123,"start_character":0,"end_line":123,"end_character":5},"in_reply_to":"dab17558_3512330a","updated":"2016-05-13 12:48:23.000000000","message":"Done","commit_id":"adb523c952b7eeab1098bf74102c4c3948c42fb3"},{"author":{"_account_id":10607,"name":"Alexandra Settle","email":"a.settle@outlook.com","username":"asettle"},"change_message_id":"b37c3d83e66245983084970440b1983562472951","unresolved":false,"context_lines":[{"line_number":129,"context_line":"       --property hw:cpu_thread_policy\u003drequire"},{"line_number":130,"context_line":""},{"line_number":131,"context_line":"For other workloads where performance is impacted by contention for resources,"},{"line_number":132,"context_line":"non-thread siblings or non-SMT hosts should be used. To force this, run:"},{"line_number":133,"context_line":""},{"line_number":134,"context_line":".. code-block:: console"},{"line_number":135,"context_line":""}],"source_content_type":"text/x-rst","patch_set":14,"id":"dab17558_15d3af39","line":132,"range":{"start_line":132,"start_character":1,"end_line":132,"end_character":36},"updated":"2016-05-13 10:46:48.000000000","message":"use non-thread siblings or non-SMT hosts*","commit_id":"adb523c952b7eeab1098bf74102c4c3948c42fb3"},{"author":{"_account_id":15334,"name":"Stephen Finucane","display_name":"stephenfin","email":"stephenfin@redhat.com","username":"sfinucan"},"change_message_id":"a578867c6c058b7379e50290f5113c81f1be8050","unresolved":false,"context_lines":[{"line_number":129,"context_line":"       --property hw:cpu_thread_policy\u003drequire"},{"line_number":130,"context_line":""},{"line_number":131,"context_line":"For other workloads where performance is impacted by contention for resources,"},{"line_number":132,"context_line":"non-thread siblings or non-SMT hosts should be used. To force this, run:"},{"line_number":133,"context_line":""},{"line_number":134,"context_line":".. code-block:: console"},{"line_number":135,"context_line":""}],"source_content_type":"text/x-rst","patch_set":14,"id":"dab17558_77683b22","line":132,"range":{"start_line":132,"start_character":1,"end_line":132,"end_character":36},"in_reply_to":"dab17558_15d3af39","updated":"2016-05-13 12:48:23.000000000","message":"Done","commit_id":"adb523c952b7eeab1098bf74102c4c3948c42fb3"},{"author":{"_account_id":10607,"name":"Alexandra Settle","email":"a.settle@outlook.com","username":"asettle"},"change_message_id":"b37c3d83e66245983084970440b1983562472951","unresolved":false,"context_lines":[{"line_number":137,"context_line":"       --property hw:cpu_policy\u003ddedicated \\"},{"line_number":138,"context_line":"       --property hw:cpu_thread_policy\u003dseparate"},{"line_number":139,"context_line":""},{"line_number":140,"context_line":"Finally for workloads where performance is minimally impacted, thread siblings"},{"line_number":141,"context_line":"can be preferred. This is the default, but it can be set explicitly:"},{"line_number":142,"context_line":""},{"line_number":143,"context_line":".. code-block:: console"}],"source_content_type":"text/x-rst","patch_set":14,"id":"dab17558_95e79f11","line":140,"range":{"start_line":140,"start_character":0,"end_line":140,"end_character":7},"updated":"2016-05-13 10:46:48.000000000","message":"Finally,*","commit_id":"adb523c952b7eeab1098bf74102c4c3948c42fb3"},{"author":{"_account_id":15334,"name":"Stephen Finucane","display_name":"stephenfin","email":"stephenfin@redhat.com","username":"sfinucan"},"change_message_id":"a578867c6c058b7379e50290f5113c81f1be8050","unresolved":false,"context_lines":[{"line_number":137,"context_line":"       --property hw:cpu_policy\u003ddedicated \\"},{"line_number":138,"context_line":"       --property hw:cpu_thread_policy\u003dseparate"},{"line_number":139,"context_line":""},{"line_number":140,"context_line":"Finally for workloads where performance is minimally impacted, thread siblings"},{"line_number":141,"context_line":"can be preferred. This is the default, but it can be set explicitly:"},{"line_number":142,"context_line":""},{"line_number":143,"context_line":".. code-block:: console"}],"source_content_type":"text/x-rst","patch_set":14,"id":"dab17558_774f9bbe","line":140,"range":{"start_line":140,"start_character":0,"end_line":140,"end_character":7},"in_reply_to":"dab17558_95e79f11","updated":"2016-05-13 12:48:23.000000000","message":"Done","commit_id":"adb523c952b7eeab1098bf74102c4c3948c42fb3"},{"author":{"_account_id":10607,"name":"Alexandra Settle","email":"a.settle@outlook.com","username":"asettle"},"change_message_id":"b37c3d83e66245983084970440b1983562472951","unresolved":false,"context_lines":[{"line_number":138,"context_line":"       --property hw:cpu_thread_policy\u003dseparate"},{"line_number":139,"context_line":""},{"line_number":140,"context_line":"Finally for workloads where performance is minimally impacted, thread siblings"},{"line_number":141,"context_line":"can be preferred. This is the default, but it can be set explicitly:"},{"line_number":142,"context_line":""},{"line_number":143,"context_line":".. code-block:: console"},{"line_number":144,"context_line":""}],"source_content_type":"text/x-rst","patch_set":14,"id":"dab17558_15ba8fe1","line":141,"range":{"start_line":141,"start_character":1,"end_line":141,"end_character":17},"updated":"2016-05-13 10:46:48.000000000","message":"is preferred? Or is optional? You can\u0027t really have can be preferred.","commit_id":"adb523c952b7eeab1098bf74102c4c3948c42fb3"},{"author":{"_account_id":15334,"name":"Stephen Finucane","display_name":"stephenfin","email":"stephenfin@redhat.com","username":"sfinucan"},"change_message_id":"a578867c6c058b7379e50290f5113c81f1be8050","unresolved":false,"context_lines":[{"line_number":138,"context_line":"       --property hw:cpu_thread_policy\u003dseparate"},{"line_number":139,"context_line":""},{"line_number":140,"context_line":"Finally for workloads where performance is minimally impacted, thread siblings"},{"line_number":141,"context_line":"can be preferred. This is the default, but it can be set explicitly:"},{"line_number":142,"context_line":""},{"line_number":143,"context_line":".. code-block:: console"},{"line_number":144,"context_line":""}],"source_content_type":"text/x-rst","patch_set":14,"id":"dab17558_375d137a","line":141,"range":{"start_line":141,"start_character":1,"end_line":141,"end_character":17},"in_reply_to":"dab17558_15ba8fe1","updated":"2016-05-13 12:48:23.000000000","message":"Done","commit_id":"adb523c952b7eeab1098bf74102c4c3948c42fb3"},{"author":{"_account_id":10607,"name":"Alexandra Settle","email":"a.settle@outlook.com","username":"asettle"},"change_message_id":"93fcb9e48143b05ec455dd7fbc54b007bbe7ed7c","unresolved":false,"context_lines":[{"line_number":21,"context_line":"found in many multi-socket systems. In a NUMA system, system memory is divided"},{"line_number":22,"context_line":"into cells or nodes that are associated with particular CPUs. Requests for"},{"line_number":23,"context_line":"memory on other nodes are possible through an interconnect bus, however,"},{"line_number":24,"context_line":"bandwidth across this shared bus is limited and, therefore, competition for"},{"line_number":25,"context_line":"this resource can incur performance penalties."},{"line_number":26,"context_line":""},{"line_number":27,"context_line":"Simultaneous Multi-Threading (SMT), known as as Hyper-Threading on Intel"}],"source_content_type":"text/x-rst","patch_set":17,"id":"dab17558_a3f500ee","line":24,"updated":"2016-05-16 10:31:16.000000000","message":"I\u0027d suggest a full stop after \u0027limited\u0027, it is very long. \"As a result, competition for this resource...\"","commit_id":"09020a6e0377cec9381281b991572ae73618b6eb"},{"author":{"_account_id":6772,"name":"Stephen Gordon","email":"sgordon@redhat.com","username":"sgordon"},"change_message_id":"20c02a0a468af007d3019048da4617d890475572","unresolved":false,"context_lines":[{"line_number":37,"context_line":".. note::"},{"line_number":38,"context_line":""},{"line_number":39,"context_line":"   The functionality described below is currently only supported by the"},{"line_number":40,"context_line":"   libvirt driver."},{"line_number":41,"context_line":""},{"line_number":42,"context_line":"When running workloads on NUMA hosts, it is important that the vCPUs executing"},{"line_number":43,"context_line":"processes are on the same NUMA node as the memory used by these processes."}],"source_content_type":"text/x-rst","patch_set":17,"id":"dab17558_85811792","line":40,"updated":"2016-05-13 17:39:34.000000000","message":"Libvirt/KVM driver correct?","commit_id":"09020a6e0377cec9381281b991572ae73618b6eb"},{"author":{"_account_id":15334,"name":"Stephen Finucane","display_name":"stephenfin","email":"stephenfin@redhat.com","username":"sfinucan"},"change_message_id":"12d2d8f65271f46920b4f304d0908bd20ac92166","unresolved":false,"context_lines":[{"line_number":37,"context_line":".. note::"},{"line_number":38,"context_line":""},{"line_number":39,"context_line":"   The functionality described below is currently only supported by the"},{"line_number":40,"context_line":"   libvirt driver."},{"line_number":41,"context_line":""},{"line_number":42,"context_line":"When running workloads on NUMA hosts, it is important that the vCPUs executing"},{"line_number":43,"context_line":"processes are on the same NUMA node as the memory used by these processes."}],"source_content_type":"text/x-rst","patch_set":17,"id":"dab17558_d4528566","line":40,"in_reply_to":"dab17558_85811792","updated":"2016-05-16 08:50:44.000000000","message":"See PS13 reply:\n\n \u003e Based on Daniel\u0027s blog [1] it would appear they do?\n \u003e [1] https://www.berrange.com/posts/2010/02/12/controlling-guest-cpu-numa-affinity-in-libvirt-with-qemu-kvm-xen/","commit_id":"09020a6e0377cec9381281b991572ae73618b6eb"},{"author":{"_account_id":17207,"name":"Atsushi SAKAI","email":"sakaia@jp.fujitsu.com","username":"sakaia"},"change_message_id":"33ade21c6af00ca0cdda03348e034eb37bc692af","unresolved":false,"context_lines":[{"line_number":37,"context_line":".. note::"},{"line_number":38,"context_line":""},{"line_number":39,"context_line":"   The functionality described below is currently only supported by the"},{"line_number":40,"context_line":"   libvirt driver."},{"line_number":41,"context_line":""},{"line_number":42,"context_line":"When running workloads on NUMA hosts, it is important that the vCPUs executing"},{"line_number":43,"context_line":"processes are on the same NUMA node as the memory used by these processes."}],"source_content_type":"text/x-rst","patch_set":17,"id":"dab17558_f5118253","line":40,"in_reply_to":"dab17558_d4528566","updated":"2016-05-16 09:58:48.000000000","message":"From seeing following code and line.\nIt supports QEMU/KVM only.\n\nlibvirt: fix disablement of NUMA \u0026 hugepages on unsupported platforms\nhttps://github.com/openstack/nova/commit/945ab28df04e22f3c1b8948972f538ee6b5e7410#diff-f4019782d93a196a0d026479e6aa61b1R4726\n\ndef _has_numa_support()","commit_id":"09020a6e0377cec9381281b991572ae73618b6eb"},{"author":{"_account_id":15334,"name":"Stephen Finucane","display_name":"stephenfin","email":"stephenfin@redhat.com","username":"sfinucan"},"change_message_id":"773bb7498753a3e70df9a15ee06ce10b13e86ab6","unresolved":false,"context_lines":[{"line_number":37,"context_line":".. note::"},{"line_number":38,"context_line":""},{"line_number":39,"context_line":"   The functionality described below is currently only supported by the"},{"line_number":40,"context_line":"   libvirt driver."},{"line_number":41,"context_line":""},{"line_number":42,"context_line":"When running workloads on NUMA hosts, it is important that the vCPUs executing"},{"line_number":43,"context_line":"processes are on the same NUMA node as the memory used by these processes."}],"source_content_type":"text/x-rst","patch_set":17,"id":"dab17558_f52342bb","line":40,"in_reply_to":"dab17558_f5118253","updated":"2016-05-16 10:04:54.000000000","message":"\u003e From seeing following code and line.\n \u003e It supports QEMU/KVM only.\n \u003e libvirt: fix disablement of NUMA \u0026 hugepages on unsupported platforms\n \u003e https://github.com/openstack/nova/commit/945ab28df04e22f3c1b8948972f538ee6b5e7410#diff-f4019782d93a196a0d026479e6aa61b1R4726\n \u003e def _has_numa_support()\n\nGood spot. I\u0027ll update now.","commit_id":"09020a6e0377cec9381281b991572ae73618b6eb"},{"author":{"_account_id":6772,"name":"Stephen Gordon","email":"sgordon@redhat.com","username":"sgordon"},"change_message_id":"20c02a0a468af007d3019048da4617d890475572","unresolved":false,"context_lines":[{"line_number":110,"context_line":".. note::"},{"line_number":111,"context_line":""},{"line_number":112,"context_line":"   The functionality described below is currently only supported by the"},{"line_number":113,"context_line":"   libvirt driver."},{"line_number":114,"context_line":""},{"line_number":115,"context_line":"By default, instance vCPU processes are not assigned to any particular host"},{"line_number":116,"context_line":"CPU, instead, they float across host CPUs like any other process. This allows"}],"source_content_type":"text/x-rst","patch_set":17,"id":"dab17558_657cf3b5","line":113,"updated":"2016-05-13 17:39:34.000000000","message":"As above.","commit_id":"09020a6e0377cec9381281b991572ae73618b6eb"},{"author":{"_account_id":17973,"name":"Ludovic Beliveau","email":"lbeliveau@gmail.com","username":"lbeliveau"},"change_message_id":"ad3fb2c02310a07e028aeb11c15a5e362821b66b","unresolved":false,"context_lines":[{"line_number":51,"context_line":""},{"line_number":52,"context_line":"By default, an instance will float across all NUMA nodes on a host. NUMA"},{"line_number":53,"context_line":"awareness can be enabled implicitly, through use of hugepages or pinned CPUs,"},{"line_number":54,"context_line":"or explicitly, through use flavor extra specs or image metadata. In all cases,"},{"line_number":55,"context_line":"the ``NUMATopologyFilter`` filter must be enabled. Details on this filter are"},{"line_number":56,"context_line":"provided in `Scheduling`_ configuration guide."},{"line_number":57,"context_line":""}],"source_content_type":"text/x-rst","patch_set":19,"id":"dab17558_6645d341","line":54,"updated":"2016-05-16 15:30:57.000000000","message":"s/through use flavor/through the use of flavor/","commit_id":"7e5f518c7fa212ff73654094e617ceeaee1957d2"}]}