)]}'
{"specs/stein/approved/overcommit-pcpu.rst":[{"author":{"_account_id":11604,"name":"sean mooney","email":"smooney@redhat.com","username":"sean-k-mooney"},"change_message_id":"1d3bc7e9e1431c56b81531a981b63c725bc36373","unresolved":false,"context_lines":[{"line_number":11,"context_line":""},{"line_number":12,"context_line":"https://blueprints.launchpad.net/nova/+spec/overcommit-pcpu"},{"line_number":13,"context_line":""},{"line_number":14,"context_line":"Allow overcommit of dedicated CPU to take the advantage of low latency over"},{"line_number":15,"context_line":"shared VCPUs thereby improving overall CPU utilization of the compute host"},{"line_number":16,"context_line":"without impacting latency requirements."},{"line_number":17,"context_line":""},{"line_number":18,"context_line":"Problem description"}],"source_content_type":"text/x-rst","patch_set":1,"id":"3f79a3b5_8e206cfa","line":15,"range":{"start_line":14,"start_character":60,"end_line":15,"end_character":12},"updated":"2018-10-23 22:08:16.000000000","message":"do you have an data to quantify this?\n\ni dont think there is nessicarrly any lower latency.\n\nthe kernel will not move your vm to a different core even though its allowed to in the free casess on less there is contention.\n\nby not allowing the cores to float and allowing over commit you could easily reduce perfromace also.","commit_id":"61393f6627e7bcedb7f281bc4abe9ffdf5731066"},{"author":{"_account_id":1011,"name":"Tushar Patil","email":"tushar.vitthal.patil@gmail.com","username":"tpatil"},"change_message_id":"afc78a7dd1a352e7c499c46c58c4b166f052d54d","unresolved":false,"context_lines":[{"line_number":11,"context_line":""},{"line_number":12,"context_line":"https://blueprints.launchpad.net/nova/+spec/overcommit-pcpu"},{"line_number":13,"context_line":""},{"line_number":14,"context_line":"Allow overcommit of dedicated CPU to take the advantage of low latency over"},{"line_number":15,"context_line":"shared VCPUs thereby improving overall CPU utilization of the compute host"},{"line_number":16,"context_line":"without impacting latency requirements."},{"line_number":17,"context_line":""},{"line_number":18,"context_line":"Problem description"}],"source_content_type":"text/x-rst","patch_set":1,"id":"3f79a3b5_b27b57f1","line":15,"range":{"start_line":14,"start_character":60,"end_line":15,"end_character":12},"in_reply_to":"3f79a3b5_8e206cfa","updated":"2018-11-21 03:54:12.000000000","message":"We have measured jitter and throughput as described here : https://www.slideshare.net/secret/BOrNLmWKU1KDma","commit_id":"61393f6627e7bcedb7f281bc4abe9ffdf5731066"},{"author":{"_account_id":11604,"name":"sean mooney","email":"smooney@redhat.com","username":"sean-k-mooney"},"change_message_id":"1d3bc7e9e1431c56b81531a981b63c725bc36373","unresolved":false,"context_lines":[{"line_number":20,"context_line":""},{"line_number":21,"context_line":"For simple network design, one vCPE subscribes one access line. This"},{"line_number":22,"context_line":"means service provider will have to deploy as many vCPE instances as"},{"line_number":23,"context_line":"accesslines (it may be over 1 million!!!). We need to pin forwarding plane"},{"line_number":24,"context_line":"process to a dedicated cpus so that packets are processed in low latency."},{"line_number":25,"context_line":"In certain cases, vCPE doesn\u0027t consume high CPU resources so assigning entire"},{"line_number":26,"context_line":"dedicated CPU is waste of CPU resource. If we allow overcommit of dedicated"},{"line_number":27,"context_line":"CPU it can accomodate more VMs on the compute host without impacting"}],"source_content_type":"text/x-rst","patch_set":1,"id":"3f79a3b5_ce6984f7","line":24,"range":{"start_line":23,"start_character":43,"end_line":24,"end_character":73},"updated":"2018-10-23 22:08:16.000000000","message":"this implies a soft realtime constraint that would likely be vilated due to context swtich resulting from over subsciption. is 1:1 cpu pinning actully required or would numa affinity be enough e.g. hw:numa_nodes\u003d1","commit_id":"61393f6627e7bcedb7f281bc4abe9ffdf5731066"},{"author":{"_account_id":1011,"name":"Tushar Patil","email":"tushar.vitthal.patil@gmail.com","username":"tpatil"},"change_message_id":"afc78a7dd1a352e7c499c46c58c4b166f052d54d","unresolved":false,"context_lines":[{"line_number":20,"context_line":""},{"line_number":21,"context_line":"For simple network design, one vCPE subscribes one access line. This"},{"line_number":22,"context_line":"means service provider will have to deploy as many vCPE instances as"},{"line_number":23,"context_line":"accesslines (it may be over 1 million!!!). We need to pin forwarding plane"},{"line_number":24,"context_line":"process to a dedicated cpus so that packets are processed in low latency."},{"line_number":25,"context_line":"In certain cases, vCPE doesn\u0027t consume high CPU resources so assigning entire"},{"line_number":26,"context_line":"dedicated CPU is waste of CPU resource. If we allow overcommit of dedicated"},{"line_number":27,"context_line":"CPU it can accomodate more VMs on the compute host without impacting"}],"source_content_type":"text/x-rst","patch_set":1,"id":"3f79a3b5_f25a2f83","line":24,"range":{"start_line":23,"start_character":43,"end_line":24,"end_character":73},"in_reply_to":"3f79a3b5_ce6984f7","updated":"2018-11-21 03:54:12.000000000","message":"For certain VNF applications we need 1:1 cpu pinning but this specs is about allowing overcommit dedicated CPU for other VNF applications which requires moderate performance.\n\nMaybe I will can remove this line if there is any confusion.","commit_id":"61393f6627e7bcedb7f281bc4abe9ffdf5731066"},{"author":{"_account_id":11604,"name":"sean mooney","email":"smooney@redhat.com","username":"sean-k-mooney"},"change_message_id":"1d3bc7e9e1431c56b81531a981b63c725bc36373","unresolved":false,"context_lines":[{"line_number":34,"context_line":"**dedicated CPU**"},{"line_number":35,"context_line":"    A physical processor that has been marked to be used for a single guest"},{"line_number":36,"context_line":"    only"},{"line_number":37,"context_line":"**overcommit dedicated CPU**"},{"line_number":38,"context_line":"    A physical processor that has been marked to be used by more than single"},{"line_number":39,"context_line":"    guest."},{"line_number":40,"context_line":""}],"source_content_type":"text/x-rst","patch_set":1,"id":"3f79a3b5_ae0be82a","line":37,"range":{"start_line":37,"start_character":13,"end_line":37,"end_character":22},"updated":"2018-10-23 22:08:16.000000000","message":"dedicated by deffinition means the core is not shared.\n\nif you were to introduce a new cpu policy\n\nhw:cpu_policy\u003dpinned that is allocated from the CPU_SHARED_SET but is pinned i would be more ok with this idea but im stronly against dedicated not menaing dedicated based on a config value i as an end user cannot see.","commit_id":"61393f6627e7bcedb7f281bc4abe9ffdf5731066"},{"author":{"_account_id":1011,"name":"Tushar Patil","email":"tushar.vitthal.patil@gmail.com","username":"tpatil"},"change_message_id":"afc78a7dd1a352e7c499c46c58c4b166f052d54d","unresolved":false,"context_lines":[{"line_number":34,"context_line":"**dedicated CPU**"},{"line_number":35,"context_line":"    A physical processor that has been marked to be used for a single guest"},{"line_number":36,"context_line":"    only"},{"line_number":37,"context_line":"**overcommit dedicated CPU**"},{"line_number":38,"context_line":"    A physical processor that has been marked to be used by more than single"},{"line_number":39,"context_line":"    guest."},{"line_number":40,"context_line":""}],"source_content_type":"text/x-rst","patch_set":1,"id":"3f79a3b5_cd52f811","line":37,"range":{"start_line":37,"start_character":13,"end_line":37,"end_character":22},"in_reply_to":"3f79a3b5_a19c59be","updated":"2018-11-21 03:54:12.000000000","message":"I\u0027m thinking of using CPU_DEDICATED_SET with the introduction of new config option `CPU_PINNING_ALLOCATION_RATIO` and add logic in Libvirt driver and NUMACell object class to manage overcommit dedicated cpus.\n\nNUMACell will store dedicated cpus in `pcpuset` attribute.  All assigned dedicated cpus will be stored in `pinned_cpus`. Right now, type of these fields is SetOfIntegersField. If we change it to \"ListOfIntegersField then it\u0027s possible to store the duplicate cpus as well.\n\nWhen NumaCell will be populated for overcommit cpus. this is how it will looks like.\n\nSay `CPU_DEDICATED_SET` \u003d [0,1,2,3]\nCPU_PINNING_ALLOCATION_RATIO \u003d 2\n\nNumaCell.pcpuset \u003d [0,0,1,1,2,2,3,3]\n\nAfter you assign any over commit dedicated cpus then it will be stored in pinned_cpus.\n\nlike this:\n\npinned_cpus \u003d [0,0]\n\nThe logic to request over commit dedicated cpus remains unchanged. For differentiating dedicated cpus compute nodes with overcommit, operator will add `overcommit_dedicated_cpu` metadata in host aggregate and each compute host belonging to the aggregate supporting overcommit dedicated cpus.","commit_id":"61393f6627e7bcedb7f281bc4abe9ffdf5731066"},{"author":{"_account_id":11604,"name":"sean mooney","email":"smooney@redhat.com","username":"sean-k-mooney"},"change_message_id":"f3208db59091f86520a0254088314be07249a219","unresolved":false,"context_lines":[{"line_number":34,"context_line":"**dedicated CPU**"},{"line_number":35,"context_line":"    A physical processor that has been marked to be used for a single guest"},{"line_number":36,"context_line":"    only"},{"line_number":37,"context_line":"**overcommit dedicated CPU**"},{"line_number":38,"context_line":"    A physical processor that has been marked to be used by more than single"},{"line_number":39,"context_line":"    guest."},{"line_number":40,"context_line":""}],"source_content_type":"text/x-rst","patch_set":1,"id":"3f79a3b5_a19c59be","line":37,"range":{"start_line":37,"start_character":13,"end_line":37,"end_character":22},"in_reply_to":"3f79a3b5_ae0be82a","updated":"2018-10-23 22:18:01.000000000","message":"just to to elablerate on this a little more \n\nmy suggestion is that hw:cpu_policy\u003dpinned would meen\ndo a 1:1 pinning of guest vcpus to cpus form the CPU_SHARED_SET with the normal oversubscition that implies.\n\nand leave hw:cpu_policy\u003ddedicated unchanged.\n\nthere would be no addtion config option just a new extra sepc \nfor the existing hw:cpu_policy. from a placment point of view hw:cpu_policy\u003dpinned instance would consume resouces from the VCPU inventory not the PCPU inventory as these are really just shared cpus wherethe vm just does not float.\n\nso with this approch no change are needed to placement or the nova configs and all that is required is a minor chage to \nthe libvirt virt driver and trackking the shared cores in the numatopology blob the same way we track dedicated cores.","commit_id":"61393f6627e7bcedb7f281bc4abe9ffdf5731066"},{"author":{"_account_id":1011,"name":"Tushar Patil","email":"tushar.vitthal.patil@gmail.com","username":"tpatil"},"change_message_id":"7c33eee9a41b90f43e6df22a2f927faa4e091cdf","unresolved":false,"context_lines":[{"line_number":34,"context_line":"**dedicated CPU**"},{"line_number":35,"context_line":"    A physical processor that has been marked to be used for a single guest"},{"line_number":36,"context_line":"    only"},{"line_number":37,"context_line":"**overcommit dedicated CPU**"},{"line_number":38,"context_line":"    A physical processor that has been marked to be used by more than single"},{"line_number":39,"context_line":"    guest."},{"line_number":40,"context_line":""}],"source_content_type":"text/x-rst","patch_set":1,"id":"3f79a3b5_cdcc3835","line":37,"range":{"start_line":37,"start_character":13,"end_line":37,"end_character":22},"in_reply_to":"3f79a3b5_cd52f811","updated":"2018-11-21 04:22:32.000000000","message":"In my last comment, I meant to say :\nFor differentiating dedicated cpus compute nodes with overcommit, operator will add `overcommit_dedicated_cpu` metadata in host aggregate and flavor extra specs.","commit_id":"61393f6627e7bcedb7f281bc4abe9ffdf5731066"},{"author":{"_account_id":8768,"name":"Chris Friesen","email":"chris.friesen@windriver.com","username":"cbf123"},"change_message_id":"fe9014b2e148640ede8a8fff3ba1a8e9218f5af0","unresolved":false,"context_lines":[{"line_number":46,"context_line":"CPUs in order to improve overall CPU utilization of the compute host without"},{"line_number":47,"context_line":"impacting latency requirements of guests used by vCPE."},{"line_number":48,"context_line":""},{"line_number":49,"context_line":"We have conducted a performance test to prove this point and you can see the"},{"line_number":50,"context_line":"jitter and throughput results published here: `POC performance results`_."},{"line_number":51,"context_line":""},{"line_number":52,"context_line":"Proposed change"},{"line_number":53,"context_line":"\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d"}],"source_content_type":"text/x-rst","patch_set":1,"id":"3f79a3b5_810b5d20","line":50,"range":{"start_line":49,"start_character":0,"end_line":50,"end_character":73},"updated":"2018-10-23 22:07:33.000000000","message":"Did you try testing using hugepages to ensure that you got numa-affinity in the \"shared\" case?\n\nWhat do you think explains the performance difference?","commit_id":"61393f6627e7bcedb7f281bc4abe9ffdf5731066"},{"author":{"_account_id":11604,"name":"sean mooney","email":"smooney@redhat.com","username":"sean-k-mooney"},"change_message_id":"81bf150fde229ffd8f7caeb116e20da261e51eee","unresolved":false,"context_lines":[{"line_number":46,"context_line":"CPUs in order to improve overall CPU utilization of the compute host without"},{"line_number":47,"context_line":"impacting latency requirements of guests used by vCPE."},{"line_number":48,"context_line":""},{"line_number":49,"context_line":"We have conducted a performance test to prove this point and you can see the"},{"line_number":50,"context_line":"jitter and throughput results published here: `POC performance results`_."},{"line_number":51,"context_line":""},{"line_number":52,"context_line":"Proposed change"},{"line_number":53,"context_line":"\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d"}],"source_content_type":"text/x-rst","patch_set":1,"id":"3f79a3b5_59ca1f47","line":50,"range":{"start_line":49,"start_character":0,"end_line":50,"end_character":73},"in_reply_to":"3f79a3b5_810b5d20","updated":"2018-11-21 15:38:25.000000000","message":"setting hw:numa_nodes\u003d1 woudl be a better\ntest as hugepages would improver perfromace in general\n\nenableing cpu pinning auto creates a single numa_node topology so testing with hw:numa_nodes\u003d1 woudl allow you to determin what improvement was due to numa affineity and what was due to pinning.","commit_id":"61393f6627e7bcedb7f281bc4abe9ffdf5731066"},{"author":{"_account_id":4690,"name":"melanie witt","display_name":"melwitt","email":"melwittt@gmail.com","username":"melwitt"},"change_message_id":"1a70f49cbcbc11cd19d20cd10b2c8f6534b0531f","unresolved":false,"context_lines":[{"line_number":58,"context_line":"assign both ``dedicated CPU`` and ``shared CPU`` to the guests."},{"line_number":59,"context_line":""},{"line_number":60,"context_line":"We would like to add ``CPU_PINNING_ALLOCATION_RATIO`` config option which will"},{"line_number":61,"context_line":"be used to report inventory for `PCPU` resource. In"},{"line_number":62,"context_line":"`Standardize CPU resource tracking`_ specs, `PCPU` inventory is created with"},{"line_number":63,"context_line":"hard coded allocation ratio as 1. Now, with ``CPU_PINNING_ALLOCATION_RATIO``,"},{"line_number":64,"context_line":"operator can decide whether a compute host should be configured for"},{"line_number":65,"context_line":"`dedicated CPU` or `overcommit dedicated CPU`."},{"line_number":66,"context_line":""},{"line_number":67,"context_line":"The operator would set the CONF options for `overcommit dedicated CPU` as"},{"line_number":68,"context_line":"below::"}],"source_content_type":"text/x-rst","patch_set":1,"id":"3f79a3b5_0bc8ce1e","line":65,"range":{"start_line":61,"start_character":49,"end_line":65,"end_character":46},"updated":"2018-10-23 21:19:03.000000000","message":"This says PCPU inventory is created in placement with allocation ratio of 1.0. So I\u0027m wondering, why do we need to add a config option for the PCPU allocation ratio? Is it not sufficient to be able to set the allocation ratio for the inventory via the placement API?","commit_id":"61393f6627e7bcedb7f281bc4abe9ffdf5731066"},{"author":{"_account_id":1011,"name":"Tushar Patil","email":"tushar.vitthal.patil@gmail.com","username":"tpatil"},"change_message_id":"afc78a7dd1a352e7c499c46c58c4b166f052d54d","unresolved":false,"context_lines":[{"line_number":58,"context_line":"assign both ``dedicated CPU`` and ``shared CPU`` to the guests."},{"line_number":59,"context_line":""},{"line_number":60,"context_line":"We would like to add ``CPU_PINNING_ALLOCATION_RATIO`` config option which will"},{"line_number":61,"context_line":"be used to report inventory for `PCPU` resource. In"},{"line_number":62,"context_line":"`Standardize CPU resource tracking`_ specs, `PCPU` inventory is created with"},{"line_number":63,"context_line":"hard coded allocation ratio as 1. Now, with ``CPU_PINNING_ALLOCATION_RATIO``,"},{"line_number":64,"context_line":"operator can decide whether a compute host should be configured for"},{"line_number":65,"context_line":"`dedicated CPU` or `overcommit dedicated CPU`."},{"line_number":66,"context_line":""},{"line_number":67,"context_line":"The operator would set the CONF options for `overcommit dedicated CPU` as"},{"line_number":68,"context_line":"below::"}],"source_content_type":"text/x-rst","patch_set":1,"id":"3f79a3b5_cd23d856","line":65,"range":{"start_line":61,"start_character":49,"end_line":65,"end_character":46},"in_reply_to":"3f79a3b5_0bc8ce1e","updated":"2018-11-21 03:54:12.000000000","message":"For managing overcommit dedicated cpus, NumaCell should be aware of allocation ratio becoz at the time of reporting compute node inventory, it\u0027s going to set `pcpuset` of NumaCell accordingly.\n\nI have explained this point in reply to Sean above. Please refer to #line 37.","commit_id":"61393f6627e7bcedb7f281bc4abe9ffdf5731066"},{"author":{"_account_id":11604,"name":"sean mooney","email":"smooney@redhat.com","username":"sean-k-mooney"},"change_message_id":"91e236894371f44da43fbaa7463f35f65ed5ead5","unresolved":false,"context_lines":[{"line_number":126,"context_line":"    Guest1"},{"line_number":127,"context_line":"        \u003cvcpu placement\u003d\u0027static\u0027\u003e6\u003c/vcpu\u003e"},{"line_number":128,"context_line":"            \u003ccputune\u003e"},{"line_number":129,"context_line":"                \u003cshares\u003e6144\u003c/shares\u003e"},{"line_number":130,"context_line":"                    \u003cvcpupin vcpu\u003d\u00270\u0027 cpuset\u003d\u00272\u0027/\u003e \u003c- Overcommit dedicated cpu"},{"line_number":131,"context_line":"                    ..."},{"line_number":132,"context_line":"            \u003c/cputune\u003e"}],"source_content_type":"text/x-rst","patch_set":1,"id":"3f79a3b5_216fe961","line":129,"range":{"start_line":129,"start_character":16,"end_line":129,"end_character":37},"updated":"2018-10-23 22:40:03.000000000","message":"the cpu shares stuff as far as i know does not really work that well for qemu/kvm. or rather its not that accurate.\nif we were to allow over subsction i would sugges not using cpu shares at all and only supporting integer allocation ratios","commit_id":"61393f6627e7bcedb7f281bc4abe9ffdf5731066"},{"author":{"_account_id":1011,"name":"Tushar Patil","email":"tushar.vitthal.patil@gmail.com","username":"tpatil"},"change_message_id":"afc78a7dd1a352e7c499c46c58c4b166f052d54d","unresolved":false,"context_lines":[{"line_number":126,"context_line":"    Guest1"},{"line_number":127,"context_line":"        \u003cvcpu placement\u003d\u0027static\u0027\u003e6\u003c/vcpu\u003e"},{"line_number":128,"context_line":"            \u003ccputune\u003e"},{"line_number":129,"context_line":"                \u003cshares\u003e6144\u003c/shares\u003e"},{"line_number":130,"context_line":"                    \u003cvcpupin vcpu\u003d\u00270\u0027 cpuset\u003d\u00272\u0027/\u003e \u003c- Overcommit dedicated cpu"},{"line_number":131,"context_line":"                    ..."},{"line_number":132,"context_line":"            \u003c/cputune\u003e"}],"source_content_type":"text/x-rst","patch_set":1,"id":"3f79a3b5_12ba4b6f","line":129,"range":{"start_line":129,"start_character":16,"end_line":129,"end_character":37},"in_reply_to":"3f79a3b5_216fe961","updated":"2018-11-21 03:54:12.000000000","message":"I had no intention to describe it here. Will remove it in the next PS.","commit_id":"61393f6627e7bcedb7f281bc4abe9ffdf5731066"},{"author":{"_account_id":4690,"name":"melanie witt","display_name":"melwitt","email":"melwittt@gmail.com","username":"melwitt"},"change_message_id":"1a70f49cbcbc11cd19d20cd10b2c8f6534b0531f","unresolved":false,"context_lines":[{"line_number":142,"context_line":"            \u003c/cputune\u003e"},{"line_number":143,"context_line":"        \u003c/vcpu\u003e"},{"line_number":144,"context_line":""},{"line_number":145,"context_line":"We will need to update NumaCell object and virt/hardware.py to accomodate"},{"line_number":146,"context_line":"``overcommit dedicated cpu`` changes."},{"line_number":147,"context_line":""},{"line_number":148,"context_line":""},{"line_number":149,"context_line":"Alternatives"}],"source_content_type":"text/x-rst","patch_set":1,"id":"3f79a3b5_6b5b8254","line":146,"range":{"start_line":145,"start_character":0,"end_line":146,"end_character":37},"updated":"2018-10-23 21:19:03.000000000","message":"Maybe I\u0027m just lacking knowledge here, but in what way do the NumaCell object and virt/hardware.py need to be updated to accomodate overcommit_dedicated_cpu? Are changes needed to add anything to the guest XML?","commit_id":"61393f6627e7bcedb7f281bc4abe9ffdf5731066"},{"author":{"_account_id":11604,"name":"sean mooney","email":"smooney@redhat.com","username":"sean-k-mooney"},"change_message_id":"91e236894371f44da43fbaa7463f35f65ed5ead5","unresolved":false,"context_lines":[{"line_number":142,"context_line":"            \u003c/cputune\u003e"},{"line_number":143,"context_line":"        \u003c/vcpu\u003e"},{"line_number":144,"context_line":""},{"line_number":145,"context_line":"We will need to update NumaCell object and virt/hardware.py to accomodate"},{"line_number":146,"context_line":"``overcommit dedicated cpu`` changes."},{"line_number":147,"context_line":""},{"line_number":148,"context_line":""},{"line_number":149,"context_line":"Alternatives"}],"source_content_type":"text/x-rst","patch_set":1,"id":"3f79a3b5_a1c3f98b","line":146,"range":{"start_line":145,"start_character":0,"end_line":146,"end_character":37},"in_reply_to":"3f79a3b5_6b5b8254","updated":"2018-10-23 22:40:03.000000000","message":"they would need to be modifed so that instaed fo just have a set of used/free cores in the blob we would need to track how many users there was of each core so you can calulate over subsctiption.\n\nalso we would need to implement teantat aware schduling to not weaken security which would require addtional info to be stored in the blob. see my comment on line 214.\n\nas for the xml no the only thig that need to be added to the xml is the vcpupin element we already generate for dedicated cpus. all that really changes is how do we calculate the set of cpus to pin to.","commit_id":"61393f6627e7bcedb7f281bc4abe9ffdf5731066"},{"author":{"_account_id":1011,"name":"Tushar Patil","email":"tushar.vitthal.patil@gmail.com","username":"tpatil"},"change_message_id":"afc78a7dd1a352e7c499c46c58c4b166f052d54d","unresolved":false,"context_lines":[{"line_number":142,"context_line":"            \u003c/cputune\u003e"},{"line_number":143,"context_line":"        \u003c/vcpu\u003e"},{"line_number":144,"context_line":""},{"line_number":145,"context_line":"We will need to update NumaCell object and virt/hardware.py to accomodate"},{"line_number":146,"context_line":"``overcommit dedicated cpu`` changes."},{"line_number":147,"context_line":""},{"line_number":148,"context_line":""},{"line_number":149,"context_line":"Alternatives"}],"source_content_type":"text/x-rst","patch_set":1,"id":"3f79a3b5_cd7f5814","line":146,"range":{"start_line":145,"start_character":0,"end_line":146,"end_character":37},"in_reply_to":"3f79a3b5_a1c3f98b","updated":"2018-11-21 03:54:12.000000000","message":"I will describe what all changes we will do in the NumaCell object and virt/hardware in detail in the next PS.","commit_id":"61393f6627e7bcedb7f281bc4abe9ffdf5731066"},{"author":{"_account_id":4690,"name":"melanie witt","display_name":"melwitt","email":"melwittt@gmail.com","username":"melwitt"},"change_message_id":"1a70f49cbcbc11cd19d20cd10b2c8f6534b0531f","unresolved":false,"context_lines":[{"line_number":200,"context_line":"Data model impact"},{"line_number":201,"context_line":"-----------------"},{"line_number":202,"context_line":""},{"line_number":203,"context_line":"The ``NUMACell`` object would need to be updated for management of overcommit"},{"line_number":204,"context_line":"of dedicated cpus."},{"line_number":205,"context_line":""},{"line_number":206,"context_line":"REST API impact"}],"source_content_type":"text/x-rst","patch_set":1,"id":"3f79a3b5_ab32fa0c","line":203,"range":{"start_line":203,"start_character":53,"end_line":203,"end_character":63},"updated":"2018-10-23 21:19:03.000000000","message":"Can a little more detail be added here to explain what is meant by \"management of overcommit\"?","commit_id":"61393f6627e7bcedb7f281bc4abe9ffdf5731066"},{"author":{"_account_id":1011,"name":"Tushar Patil","email":"tushar.vitthal.patil@gmail.com","username":"tpatil"},"change_message_id":"afc78a7dd1a352e7c499c46c58c4b166f052d54d","unresolved":false,"context_lines":[{"line_number":200,"context_line":"Data model impact"},{"line_number":201,"context_line":"-----------------"},{"line_number":202,"context_line":""},{"line_number":203,"context_line":"The ``NUMACell`` object would need to be updated for management of overcommit"},{"line_number":204,"context_line":"of dedicated cpus."},{"line_number":205,"context_line":""},{"line_number":206,"context_line":"REST API impact"}],"source_content_type":"text/x-rst","patch_set":1,"id":"3f79a3b5_0db470d2","line":203,"range":{"start_line":203,"start_character":53,"end_line":203,"end_character":63},"in_reply_to":"3f79a3b5_ab32fa0c","updated":"2018-11-21 03:54:12.000000000","message":"I will add NumaCell class changes in the next PS.","commit_id":"61393f6627e7bcedb7f281bc4abe9ffdf5731066"},{"author":{"_account_id":11604,"name":"sean mooney","email":"smooney@redhat.com","username":"sean-k-mooney"},"change_message_id":"91e236894371f44da43fbaa7463f35f65ed5ead5","unresolved":false,"context_lines":[{"line_number":211,"context_line":"Security impact"},{"line_number":212,"context_line":"---------------"},{"line_number":213,"context_line":""},{"line_number":214,"context_line":"None"},{"line_number":215,"context_line":""},{"line_number":216,"context_line":"Notifications impact"},{"line_number":217,"context_line":"--------------------"}],"source_content_type":"text/x-rst","patch_set":1,"id":"3f79a3b5_01bd0dd6","line":214,"range":{"start_line":214,"start_character":0,"end_line":214,"end_character":4},"updated":"2018-10-23 22:40:03.000000000","message":"in a public cloud enviornment we would have to guarentee that \nto instnace form different tenatns are never pinned to the same core or hyper tread or this would grealtly simply executing a L1TF or specter attack.\n\n\neffectivly, without expcitly implementing tenant aware cpu pinning on top of what is proposed in this spec this would allow two vms from two different tenants to be pinned to\nthe same host cpu/thread. the perfromace improvement you point too in https://www.slideshare.net/secret/BOrNLmWKU1KDma\n\nis partly as a result of the fact the vms can share via l1 and l2 cache as they are running on the same core. that same sharing is what makes a vm vulnerable due to speculation when it shares a core with another vm.","commit_id":"61393f6627e7bcedb7f281bc4abe9ffdf5731066"},{"author":{"_account_id":1011,"name":"Tushar Patil","email":"tushar.vitthal.patil@gmail.com","username":"tpatil"},"change_message_id":"afc78a7dd1a352e7c499c46c58c4b166f052d54d","unresolved":false,"context_lines":[{"line_number":211,"context_line":"Security impact"},{"line_number":212,"context_line":"---------------"},{"line_number":213,"context_line":""},{"line_number":214,"context_line":"None"},{"line_number":215,"context_line":""},{"line_number":216,"context_line":"Notifications impact"},{"line_number":217,"context_line":"--------------------"}],"source_content_type":"text/x-rst","patch_set":1,"id":"3f79a3b5_0dcf908a","line":214,"range":{"start_line":214,"start_character":0,"end_line":214,"end_character":4},"in_reply_to":"3f79a3b5_01bd0dd6","updated":"2018-11-21 03:54:12.000000000","message":"I will describe this important point in the next PS. Thank you.","commit_id":"61393f6627e7bcedb7f281bc4abe9ffdf5731066"},{"author":{"_account_id":4690,"name":"melanie witt","display_name":"melwitt","email":"melwittt@gmail.com","username":"melwitt"},"change_message_id":"1a70f49cbcbc11cd19d20cd10b2c8f6534b0531f","unresolved":false,"context_lines":[{"line_number":272,"context_line":"* Create ``CONF.cpu_pinning_allocation_ratio`` config option and use this"},{"line_number":273,"context_line":"  config option to report inventory of PCPU resource"},{"line_number":274,"context_line":"* Modify ``NumaCell`` object for management of overcommit PCPUs"},{"line_number":275,"context_line":"* TBD"},{"line_number":276,"context_line":"* Unit Tests"},{"line_number":277,"context_line":"* Functional Tests"},{"line_number":278,"context_line":"* Update documentation"}],"source_content_type":"text/x-rst","patch_set":1,"id":"3f79a3b5_4b6b261b","line":275,"range":{"start_line":275,"start_character":2,"end_line":275,"end_character":5},"updated":"2018-10-23 21:19:03.000000000","message":"This got accidentally left here?","commit_id":"61393f6627e7bcedb7f281bc4abe9ffdf5731066"},{"author":{"_account_id":1011,"name":"Tushar Patil","email":"tushar.vitthal.patil@gmail.com","username":"tpatil"},"change_message_id":"afc78a7dd1a352e7c499c46c58c4b166f052d54d","unresolved":false,"context_lines":[{"line_number":272,"context_line":"* Create ``CONF.cpu_pinning_allocation_ratio`` config option and use this"},{"line_number":273,"context_line":"  config option to report inventory of PCPU resource"},{"line_number":274,"context_line":"* Modify ``NumaCell`` object for management of overcommit PCPUs"},{"line_number":275,"context_line":"* TBD"},{"line_number":276,"context_line":"* Unit Tests"},{"line_number":277,"context_line":"* Functional Tests"},{"line_number":278,"context_line":"* Update documentation"}],"source_content_type":"text/x-rst","patch_set":1,"id":"3f79a3b5_4de4a80b","line":275,"range":{"start_line":275,"start_character":2,"end_line":275,"end_character":5},"in_reply_to":"3f79a3b5_4b6b261b","updated":"2018-11-21 03:54:12.000000000","message":"My mistake. Will remove it in the next PS.","commit_id":"61393f6627e7bcedb7f281bc4abe9ffdf5731066"},{"author":{"_account_id":4690,"name":"melanie witt","display_name":"melwitt","email":"melwittt@gmail.com","username":"melwitt"},"change_message_id":"1a70f49cbcbc11cd19d20cd10b2c8f6534b0531f","unresolved":false,"context_lines":[{"line_number":275,"context_line":"* TBD"},{"line_number":276,"context_line":"* Unit Tests"},{"line_number":277,"context_line":"* Functional Tests"},{"line_number":278,"context_line":"* Update documentation"},{"line_number":279,"context_line":""},{"line_number":280,"context_line":""},{"line_number":281,"context_line":"Dependencies"}],"source_content_type":"text/x-rst","patch_set":1,"id":"3f79a3b5_0b968ef0","line":278,"updated":"2018-10-23 21:19:03.000000000","message":"It looks like some work items are missing. What about:\n\n* Add ``resources1:PCPU\u003d$AMOUNT`` extra spec\n* Add `aggregate_instance_extra_specs:overcommit_dedicated_cpu\u003d\u003cvalue\u003e` extra spec","commit_id":"61393f6627e7bcedb7f281bc4abe9ffdf5731066"},{"author":{"_account_id":1011,"name":"Tushar Patil","email":"tushar.vitthal.patil@gmail.com","username":"tpatil"},"change_message_id":"afc78a7dd1a352e7c499c46c58c4b166f052d54d","unresolved":false,"context_lines":[{"line_number":275,"context_line":"* TBD"},{"line_number":276,"context_line":"* Unit Tests"},{"line_number":277,"context_line":"* Functional Tests"},{"line_number":278,"context_line":"* Update documentation"},{"line_number":279,"context_line":""},{"line_number":280,"context_line":""},{"line_number":281,"context_line":"Dependencies"}],"source_content_type":"text/x-rst","patch_set":1,"id":"3f79a3b5_2dc28c55","line":278,"in_reply_to":"3f79a3b5_0b968ef0","updated":"2018-11-21 03:54:12.000000000","message":"1) resources1:PCPU\u003d$AMOUNT`` extra spec will be implemented in the `Standardize CPU resource tracking` specs. \n\n2) aggregate_instance_extra_specs:overcommit_dedicated_cpu\u003d\u003cvalue\u003e\nI will add this item in the next PS. Thank you for pointing it out.","commit_id":"61393f6627e7bcedb7f281bc4abe9ffdf5731066"}]}
