)]}'
{"Empty file":[{"author":{"_account_id":6772,"name":"Stephen Gordon","email":"sgordon@redhat.com","username":"sgordon"},"change_message_id":"d25049f1d0f12714fa1f02bce7e5da3f592c1b20","unresolved":false,"context_lines":[],"source_content_type":"","patch_set":4,"id":"5a890539_4f6dd297","side":"PARENT","updated":"2014-11-13 16:46:37.000000000","message":"Question asked in NFV design session was whether we really want to expose the # of queues this way. Also are we capping the # of queues at the maximum (8 or 16 - cant remember)? As it reads the number would continue to increase beyond this based on # of vCPUs?","commit_id":"08c972e0b5cc0991148d15508429103fd991de31"}],"specs/kilo/approved/libvirt-virtiomq.rst":[{"author":{"_account_id":1779,"name":"Daniel Berrange","email":"berrange@redhat.com","username":"berrange"},"change_message_id":"830c45656739ba7402d11f5d7aa5fcb2a8e4fec4","unresolved":false,"context_lines":[{"line_number":38,"context_line":"- Multiqueue virtio-net works well for incoming traffic, but can"},{"line_number":39,"context_line":"  occasionally hurt performance for outgoing traffic."},{"line_number":40,"context_line":"- Enabling multiqueue virtio-net increases the total network throughput,"},{"line_number":41,"context_line":"  and in parallel increases CPU consumption."},{"line_number":42,"context_line":""},{"line_number":43,"context_line":"Use Cases"},{"line_number":44,"context_line":"---------"}],"source_content_type":"text/x-rst","patch_set":3,"id":"9aa7fdbe_758d6e31","line":41,"updated":"2014-10-20 10:15:33.000000000","message":"The bigger issue is that a guest OS is limited to 256 MSI vectors. Each NIC queue requires an MSI vector. So if you have many virtio NICs and many vCPUs then we have the possibility of hitting the guest MSI limit.\n\nI can imagine there are times where having huge numbers of NICs is important but having multi-queue is not important. eg if each guest NIC is used by exactly one vCPU.  So this is a compelling reason to not unconditionally enable multiqueue.\n\nThe KVM maintainers also tell me that even if enabled in the host QEMU config, multi-queue does *not* get enabled in the guest OS. The guest OS administrator needs to manually turn it on for each guest NIC that wants it, using ethtool.\n\nSo the guest admin already has the ability to turn it on / off per NIC, but that would not control the MSI usage. eg if MQ was enabled in host but disabled in guest, the MSI vectors would still be consumed (wasted).","commit_id":"7b5f8a7c2a77a7408273237c140ad69e0f21166f"},{"author":{"_account_id":8802,"name":"Vladik Romanovsky","email":"vromanso@redhat.com","username":"vladikr"},"change_message_id":"dd88ba03618f08b8066dab134335224f88b4a9c6","unresolved":false,"context_lines":[{"line_number":38,"context_line":"- Multiqueue virtio-net works well for incoming traffic, but can"},{"line_number":39,"context_line":"  occasionally hurt performance for outgoing traffic."},{"line_number":40,"context_line":"- Enabling multiqueue virtio-net increases the total network throughput,"},{"line_number":41,"context_line":"  and in parallel increases CPU consumption."},{"line_number":42,"context_line":""},{"line_number":43,"context_line":"Use Cases"},{"line_number":44,"context_line":"---------"}],"source_content_type":"text/x-rst","patch_set":3,"id":"9aa7fdbe_a8ec18bc","line":41,"in_reply_to":"9aa7fdbe_758d6e31","updated":"2014-10-29 12:39:31.000000000","message":"Done","commit_id":"7b5f8a7c2a77a7408273237c140ad69e0f21166f"},{"author":{"_account_id":1779,"name":"Daniel Berrange","email":"berrange@redhat.com","username":"berrange"},"change_message_id":"9db873640ba64a2a041ed777bb0929d14f088898","unresolved":false,"context_lines":[{"line_number":39,"context_line":"  occasionally hurt performance for outgoing traffic."},{"line_number":40,"context_line":"- Enabling multiqueue virtio-net increases the total network throughput,"},{"line_number":41,"context_line":"  and in parallel increases CPU consumption."},{"line_number":42,"context_line":""},{"line_number":43,"context_line":"Use Cases"},{"line_number":44,"context_line":"---------"},{"line_number":45,"context_line":"This is useful when an instance needs to increase the network throughput"}],"source_content_type":"text/x-rst","patch_set":3,"id":"9aa7fdbe_906e60c8","line":42,"updated":"2014-10-20 10:20:40.000000000","message":"Also note that each virtio-net queue consumes 64 KB of kernel memory for the vhost driver.","commit_id":"7b5f8a7c2a77a7408273237c140ad69e0f21166f"},{"author":{"_account_id":1849,"name":"Joe Gordon","email":"joe.gordon0@gmail.com","username":"jogo"},"change_message_id":"b62a964ff734391db7279bde8c74a7799bc84b47","unresolved":false,"context_lines":[{"line_number":42,"context_line":""},{"line_number":43,"context_line":"Use Cases"},{"line_number":44,"context_line":"---------"},{"line_number":45,"context_line":"This is useful when an instance needs to increase the network throughput"},{"line_number":46,"context_line":""},{"line_number":47,"context_line":"Project Priority"},{"line_number":48,"context_line":"-----------------"}],"source_content_type":"text/x-rst","patch_set":3,"id":"9aa7fdbe_f25ee611","line":45,"updated":"2014-10-17 22:10:04.000000000","message":"useful for who? Can you add more detail here","commit_id":"7b5f8a7c2a77a7408273237c140ad69e0f21166f"},{"author":{"_account_id":8802,"name":"Vladik Romanovsky","email":"vromanso@redhat.com","username":"vladikr"},"change_message_id":"dd88ba03618f08b8066dab134335224f88b4a9c6","unresolved":false,"context_lines":[{"line_number":42,"context_line":""},{"line_number":43,"context_line":"Use Cases"},{"line_number":44,"context_line":"---------"},{"line_number":45,"context_line":"This is useful when an instance needs to increase the network throughput"},{"line_number":46,"context_line":""},{"line_number":47,"context_line":"Project Priority"},{"line_number":48,"context_line":"-----------------"}],"source_content_type":"text/x-rst","patch_set":3,"id":"9aa7fdbe_3dbbd276","line":45,"in_reply_to":"9aa7fdbe_f25ee611","updated":"2014-10-29 12:39:31.000000000","message":"Done","commit_id":"7b5f8a7c2a77a7408273237c140ad69e0f21166f"},{"author":{"_account_id":1849,"name":"Joe Gordon","email":"joe.gordon0@gmail.com","username":"jogo"},"change_message_id":"b62a964ff734391db7279bde8c74a7799bc84b47","unresolved":false,"context_lines":[{"line_number":46,"context_line":""},{"line_number":47,"context_line":"Project Priority"},{"line_number":48,"context_line":"-----------------"},{"line_number":49,"context_line":"Low"},{"line_number":50,"context_line":""},{"line_number":51,"context_line":""},{"line_number":52,"context_line":"Proposed change"}],"source_content_type":"text/x-rst","patch_set":3,"id":"9aa7fdbe_b2645ee6","line":49,"updated":"2014-10-17 22:10:04.000000000","message":"This should be None. As we don\u0027t have any set priorities right now.","commit_id":"7b5f8a7c2a77a7408273237c140ad69e0f21166f"},{"author":{"_account_id":8802,"name":"Vladik Romanovsky","email":"vromanso@redhat.com","username":"vladikr"},"change_message_id":"dd88ba03618f08b8066dab134335224f88b4a9c6","unresolved":false,"context_lines":[{"line_number":46,"context_line":""},{"line_number":47,"context_line":"Project Priority"},{"line_number":48,"context_line":"-----------------"},{"line_number":49,"context_line":"Low"},{"line_number":50,"context_line":""},{"line_number":51,"context_line":""},{"line_number":52,"context_line":"Proposed change"}],"source_content_type":"text/x-rst","patch_set":3,"id":"9aa7fdbe_48fbac78","line":49,"in_reply_to":"9aa7fdbe_b2645ee6","updated":"2014-10-29 12:39:31.000000000","message":"Done","commit_id":"7b5f8a7c2a77a7408273237c140ad69e0f21166f"},{"author":{"_account_id":1849,"name":"Joe Gordon","email":"joe.gordon0@gmail.com","username":"jogo"},"change_message_id":"1e04e916fce37eabf66444a5b94976bd0e964572","unresolved":false,"context_lines":[{"line_number":57,"context_line":"disable it, for cetrain workloads, where this feature will only lead to a"},{"line_number":58,"context_line":"reduced performance."},{"line_number":59,"context_line":""},{"line_number":60,"context_line":"Introducing a new paramenter to the flavour extra specs, for the administrator"},{"line_number":61,"context_line":"to control the virtio-net multiqueue feature."},{"line_number":62,"context_line":""},{"line_number":63,"context_line":"hw:vif_multiqueue_enabled \u003d true|false"}],"source_content_type":"text/x-rst","patch_set":3,"id":"9aa7fdbe_b2ab7ed5","line":60,"updated":"2014-10-17 22:30:10.000000000","message":"nit: to be consistent, we should use US English. so flavor.","commit_id":"7b5f8a7c2a77a7408273237c140ad69e0f21166f"},{"author":{"_account_id":8802,"name":"Vladik Romanovsky","email":"vromanso@redhat.com","username":"vladikr"},"change_message_id":"dd88ba03618f08b8066dab134335224f88b4a9c6","unresolved":false,"context_lines":[{"line_number":57,"context_line":"disable it, for cetrain workloads, where this feature will only lead to a"},{"line_number":58,"context_line":"reduced performance."},{"line_number":59,"context_line":""},{"line_number":60,"context_line":"Introducing a new paramenter to the flavour extra specs, for the administrator"},{"line_number":61,"context_line":"to control the virtio-net multiqueue feature."},{"line_number":62,"context_line":""},{"line_number":63,"context_line":"hw:vif_multiqueue_enabled \u003d true|false"}],"source_content_type":"text/x-rst","patch_set":3,"id":"9aa7fdbe_88f11497","line":60,"in_reply_to":"9aa7fdbe_b2ab7ed5","updated":"2014-10-29 12:39:31.000000000","message":"Done","commit_id":"7b5f8a7c2a77a7408273237c140ad69e0f21166f"},{"author":{"_account_id":1779,"name":"Daniel Berrange","email":"berrange@redhat.com","username":"berrange"},"change_message_id":"830c45656739ba7402d11f5d7aa5fcb2a8e4fec4","unresolved":false,"context_lines":[{"line_number":69,"context_line":"hw_vif_multiqueue_enabled \u003d true|false"},{"line_number":70,"context_line":""},{"line_number":71,"context_line":"The Libvirt driver will be enhanced to match the number of virtio-net queues to"},{"line_number":72,"context_line":"the number of VCPUs."},{"line_number":73,"context_line":""},{"line_number":74,"context_line":"Alternatives"},{"line_number":75,"context_line":"------------"}],"source_content_type":"text/x-rst","patch_set":3,"id":"9aa7fdbe_1581828f","line":72,"updated":"2014-10-20 10:15:33.000000000","message":"I\u0027m really wondering whether it is right to do it this way via flavour + image properties. In particular I\u0027m not hugely liking the idea of exposing it to image properties.  I\u0027m also concerned we might need ability to toggle it per NIC.\n\nFinally I\u0027m inclined to say that we should enable it in the host by default. Guest OS already need to enable it too, so enabling unconditionally in the host we give better out of the box config for the common use case.\n\nSo an alternative approach might\n\n - Flavor  hw:vif_multiqueue_permitted\u003dtrue|false  (default true)\n\n - Network port property to set number of queues for that port (default \u003d\u003d # of vCPUs)\n\nSo out of the box:\n\n - guests will have multiqueue with # of vCPUs.\n - Cloud admin can disable this by setting flavour property\n - End user can control # of queues when providing network port config\n - Guest OS images need to enable it via ethtool","commit_id":"7b5f8a7c2a77a7408273237c140ad69e0f21166f"},{"author":{"_account_id":8802,"name":"Vladik Romanovsky","email":"vromanso@redhat.com","username":"vladikr"},"change_message_id":"dd88ba03618f08b8066dab134335224f88b4a9c6","unresolved":false,"context_lines":[{"line_number":69,"context_line":"hw_vif_multiqueue_enabled \u003d true|false"},{"line_number":70,"context_line":""},{"line_number":71,"context_line":"The Libvirt driver will be enhanced to match the number of virtio-net queues to"},{"line_number":72,"context_line":"the number of VCPUs."},{"line_number":73,"context_line":""},{"line_number":74,"context_line":"Alternatives"},{"line_number":75,"context_line":"------------"}],"source_content_type":"text/x-rst","patch_set":3,"id":"9aa7fdbe_08352445","line":72,"in_reply_to":"9aa7fdbe_1581828f","updated":"2014-10-29 12:39:31.000000000","message":"Done","commit_id":"7b5f8a7c2a77a7408273237c140ad69e0f21166f"},{"author":{"_account_id":1849,"name":"Joe Gordon","email":"joe.gordon0@gmail.com","username":"jogo"},"change_message_id":"1e04e916fce37eabf66444a5b94976bd0e964572","unresolved":false,"context_lines":[{"line_number":140,"context_line":"Testing"},{"line_number":141,"context_line":"\u003d\u003d\u003d\u003d\u003d\u003d\u003d"},{"line_number":142,"context_line":""},{"line_number":143,"context_line":"Unit tests."},{"line_number":144,"context_line":""},{"line_number":145,"context_line":"Documentation Impact"},{"line_number":146,"context_line":"\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d"}],"source_content_type":"text/x-rst","patch_set":3,"id":"9aa7fdbe_b2e37e04","line":143,"updated":"2014-10-17 22:30:10.000000000","message":"What version of libvirt does this require","commit_id":"7b5f8a7c2a77a7408273237c140ad69e0f21166f"},{"author":{"_account_id":8802,"name":"Vladik Romanovsky","email":"vromanso@redhat.com","username":"vladikr"},"change_message_id":"dd88ba03618f08b8066dab134335224f88b4a9c6","unresolved":false,"context_lines":[{"line_number":140,"context_line":"Testing"},{"line_number":141,"context_line":"\u003d\u003d\u003d\u003d\u003d\u003d\u003d"},{"line_number":142,"context_line":""},{"line_number":143,"context_line":"Unit tests."},{"line_number":144,"context_line":""},{"line_number":145,"context_line":"Documentation Impact"},{"line_number":146,"context_line":"\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d"}],"source_content_type":"text/x-rst","patch_set":3,"id":"9aa7fdbe_5dc096e0","line":143,"in_reply_to":"9aa7fdbe_b2e37e04","updated":"2014-10-29 12:39:31.000000000","message":"Done","commit_id":"7b5f8a7c2a77a7408273237c140ad69e0f21166f"},{"author":{"_account_id":1779,"name":"Daniel Berrange","email":"berrange@redhat.com","username":"berrange"},"change_message_id":"1fb06d654a1853b7de4d839a4d367b706fa0b347","unresolved":false,"context_lines":[{"line_number":35,"context_line":"Although the virtio-net multiqueue feature provides a performance benefit,"},{"line_number":36,"context_line":"it has some limitations, therefor should not be unconditionally enabled:"},{"line_number":37,"context_line":""},{"line_number":38,"context_line":"- Guest OS is limited to 256 MSI vectors. Each NIC queue requires a MSI vector."},{"line_number":39,"context_line":"  Defining an instance with multiple virtio NICs and vCPUs might lead to a"},{"line_number":40,"context_line":"  possibility of hitting the guest MSI limit."},{"line_number":41,"context_line":"- virtio-net multiqueue works well for incoming traffic, but can"}],"source_content_type":"text/x-rst","patch_set":4,"id":"5a890539_aa99c420","line":38,"updated":"2014-11-13 17:19:47.000000000","message":"I\u0027ve just been told it is actually  ~200, not 256.  Also any virtio device  or assigned PCI device will consume them, not just NICs.","commit_id":"f425f7baba83b8eabb77e52ad1a89253d4a0733b"},{"author":{"_account_id":782,"name":"John Garbutt","email":"john@johngarbutt.com","username":"johngarbutt"},"change_message_id":"7b7d5ae0df27df15247275bc7c6e385d65ac7b8b","unresolved":false,"context_lines":[{"line_number":55,"context_line":"Use Cases"},{"line_number":56,"context_line":"---------"},{"line_number":57,"context_line":"Guest instance users may benefit from an increased network"},{"line_number":58,"context_line":"performance and throughput."},{"line_number":59,"context_line":""},{"line_number":60,"context_line":"Project Priority"},{"line_number":61,"context_line":"-----------------"}],"source_content_type":"text/x-rst","patch_set":4,"id":"5a890539_4c4760a3","line":58,"updated":"2014-12-03 17:43:55.000000000","message":"I think we need to add some specific use cases here, like the ones Daniel describes below.\n\nTalk about why you need the API, so its a finite resource on the hypervisor, but you need many NIC, but only one needs to be high performance.","commit_id":"f425f7baba83b8eabb77e52ad1a89253d4a0733b"},{"author":{"_account_id":782,"name":"John Garbutt","email":"john@johngarbutt.com","username":"johngarbutt"},"change_message_id":"7b7d5ae0df27df15247275bc7c6e385d65ac7b8b","unresolved":false,"context_lines":[{"line_number":73,"context_line":"Introducing a new parameter to the flavor extra specs, for the administrator"},{"line_number":74,"context_line":"to control the virtio-net multiqueue feature."},{"line_number":75,"context_line":""},{"line_number":76,"context_line":"    hw:vif_multiqueue_permitted\u003dtrue|false  (default true)"},{"line_number":77,"context_line":""},{"line_number":78,"context_line":"Users will be able to set the number of queues as part of the network port"},{"line_number":79,"context_line":"properties."}],"source_content_type":"text/x-rst","patch_set":4,"id":"5a890539_0cd8f8ed","line":76,"updated":"2014-12-03 17:43:55.000000000","message":"we should at least match the detail in the API below, I think, so queues\u003dvcpu or 1 or any other number, maybe?\n\nNot sure we want queues\u003dvcpu, given its in the flavor, that seems a bit odd, probably better to specify the specific number only, maybe?","commit_id":"f425f7baba83b8eabb77e52ad1a89253d4a0733b"},{"author":{"_account_id":8802,"name":"Vladik Romanovsky","email":"vromanso@redhat.com","username":"vladikr"},"change_message_id":"ca2efd5a9c9f01027b9b6a90951ae1e0b4948e3c","unresolved":false,"context_lines":[{"line_number":73,"context_line":"Introducing a new parameter to the flavor extra specs, for the administrator"},{"line_number":74,"context_line":"to control the virtio-net multiqueue feature."},{"line_number":75,"context_line":""},{"line_number":76,"context_line":"    hw:vif_multiqueue_permitted\u003dtrue|false  (default true)"},{"line_number":77,"context_line":""},{"line_number":78,"context_line":"Users will be able to set the number of queues as part of the network port"},{"line_number":79,"context_line":"properties."}],"source_content_type":"text/x-rst","patch_set":4,"id":"3a961159_7013760d","line":76,"in_reply_to":"5a890539_0cd8f8ed","updated":"2014-12-09 21:36:22.000000000","message":"Well, the overall idea in this blueprint is to have the administrator permitting this feature in extra specs. Once it\u0027s permitted, the default value of queues, for each vNIC, will be set to the number of vcpus.\nThe user will be able to override this setting per each virtual nic via the API.\n\nThe default value of queues\u003dvcpu (having a thread per vcpu) is the prefered value and provides best performance in most conditions.\nThis is according to the KVM team who introduced this feature in qemu.\n(The link is in the reference)\n\nAccording to the above and Daniel\u0027s explanation, we cannot match the extra spec setting to API, as we cannot control each vNIC separately.","commit_id":"f425f7baba83b8eabb77e52ad1a89253d4a0733b"},{"author":{"_account_id":6772,"name":"Stephen Gordon","email":"sgordon@redhat.com","username":"sgordon"},"change_message_id":"7302aa16df0164812ca8ff50cefac3ce09433888","unresolved":false,"context_lines":[{"line_number":80,"context_line":"The default number of queues will match the number of vCPUs, defined for the"},{"line_number":81,"context_line":"instance."},{"line_number":82,"context_line":""},{"line_number":83,"context_line":"    nova boot -nic -port_id -port_queues\u003d#"},{"line_number":84,"context_line":""},{"line_number":85,"context_line":"NOTE: Virtio-net multiqueue should be enabled in the guest OS manually, using"},{"line_number":86,"context_line":"ethtool."}],"source_content_type":"text/x-rst","patch_set":4,"id":"5a890539_2f2166b4","line":83,"updated":"2014-11-13 17:02:58.000000000","message":"Question raised in NFV design session was do we really want to expose # of queues? Also as written text above indicates we will always scale up with # of vCPUs but I believe there is a maximum of 8 or 16 queues?","commit_id":"f425f7baba83b8eabb77e52ad1a89253d4a0733b"},{"author":{"_account_id":1779,"name":"Daniel Berrange","email":"berrange@redhat.com","username":"berrange"},"change_message_id":"ace1941ed3709e924a13b53bfcce63743843b880","unresolved":false,"context_lines":[{"line_number":80,"context_line":"The default number of queues will match the number of vCPUs, defined for the"},{"line_number":81,"context_line":"instance."},{"line_number":82,"context_line":""},{"line_number":83,"context_line":"    nova boot -nic -port_id -port_queues\u003d#"},{"line_number":84,"context_line":""},{"line_number":85,"context_line":"NOTE: Virtio-net multiqueue should be enabled in the guest OS manually, using"},{"line_number":86,"context_line":"ethtool."}],"source_content_type":"text/x-rst","patch_set":4,"id":"5a890539_ea802c48","line":83,"in_reply_to":"5a890539_2f2166b4","updated":"2014-11-13 17:07:37.000000000","message":"If we didn\u0027t expose ability to override the queue count, and only have the hardcoded num of queues equal to number of vCPUs, we could fairly easily hit the limit when using guests with large vCPU counts and many NICs.\n\neg  a 32 vCPU guest would only be able to have 8 NICs if we always used # of queues \u003d\u003d # of vcpus","commit_id":"f425f7baba83b8eabb77e52ad1a89253d4a0733b"},{"author":{"_account_id":8802,"name":"Vladik Romanovsky","email":"vromanso@redhat.com","username":"vladikr"},"change_message_id":"0767827be0ceffd390379574718fd150d7d87190","unresolved":false,"context_lines":[{"line_number":80,"context_line":"The default number of queues will match the number of vCPUs, defined for the"},{"line_number":81,"context_line":"instance."},{"line_number":82,"context_line":""},{"line_number":83,"context_line":"    nova boot -nic -port_id -port_queues\u003d#"},{"line_number":84,"context_line":""},{"line_number":85,"context_line":"NOTE: Virtio-net multiqueue should be enabled in the guest OS manually, using"},{"line_number":86,"context_line":"ethtool."}],"source_content_type":"text/x-rst","patch_set":4,"id":"5a890539_aae1c4d1","line":83,"in_reply_to":"5a890539_2f2166b4","updated":"2014-11-13 17:14:28.000000000","message":"Thanks.\nI didn\u0027t know that 16 queues is an actual limitations, thought that this is more of a suggested top value.","commit_id":"f425f7baba83b8eabb77e52ad1a89253d4a0733b"},{"author":{"_account_id":782,"name":"John Garbutt","email":"john@johngarbutt.com","username":"johngarbutt"},"change_message_id":"7b7d5ae0df27df15247275bc7c6e385d65ac7b8b","unresolved":false,"context_lines":[{"line_number":80,"context_line":"The default number of queues will match the number of vCPUs, defined for the"},{"line_number":81,"context_line":"instance."},{"line_number":82,"context_line":""},{"line_number":83,"context_line":"    nova boot -nic -port_id -port_queues\u003d#"},{"line_number":84,"context_line":""},{"line_number":85,"context_line":"NOTE: Virtio-net multiqueue should be enabled in the guest OS manually, using"},{"line_number":86,"context_line":"ethtool."}],"source_content_type":"text/x-rst","patch_set":4,"id":"5a890539_0737f3e3","line":83,"in_reply_to":"5a890539_5d654c1c","updated":"2014-12-03 17:43:55.000000000","message":"So this really needs to be added into Neutron before we can add it into Nova, I think you will do well to split up this blueprint.","commit_id":"f425f7baba83b8eabb77e52ad1a89253d4a0733b"},{"author":{"_account_id":6598,"name":"Berezovsky Irena","email":"irenab.dev@gmail.com","username":"irenab"},"change_message_id":"3fddbf18276d3399f7a1af01afc7fc075ee7a62e","unresolved":false,"context_lines":[{"line_number":80,"context_line":"The default number of queues will match the number of vCPUs, defined for the"},{"line_number":81,"context_line":"instance."},{"line_number":82,"context_line":""},{"line_number":83,"context_line":"    nova boot -nic -port_id -port_queues\u003d#"},{"line_number":84,"context_line":""},{"line_number":85,"context_line":"NOTE: Virtio-net multiqueue should be enabled in the guest OS manually, using"},{"line_number":86,"context_line":"ethtool."}],"source_content_type":"text/x-rst","patch_set":4,"id":"5a890539_5d654c1c","line":83,"in_reply_to":"5a890539_aae1c4d1","updated":"2014-11-16 20:10:59.000000000","message":"Proposed api reminds me previous attempt to request vnic_type via nova boot. It was rejected, and main reason was that API should be abstract. NIC flavor was mentioned as a possible direction to define NIC related settings.","commit_id":"f425f7baba83b8eabb77e52ad1a89253d4a0733b"},{"author":{"_account_id":782,"name":"John Garbutt","email":"john@johngarbutt.com","username":"johngarbutt"},"change_message_id":"bf9967785b8c127fbd11e9daf8ce01ffd308d790","unresolved":false,"context_lines":[{"line_number":83,"context_line":"    nova boot -nic -port_id -port_queues\u003d#"},{"line_number":84,"context_line":""},{"line_number":85,"context_line":"NOTE: Virtio-net multiqueue should be enabled in the guest OS manually, using"},{"line_number":86,"context_line":"ethtool."},{"line_number":87,"context_line":""},{"line_number":88,"context_line":"Alternatives"},{"line_number":89,"context_line":"------------"}],"source_content_type":"text/x-rst","patch_set":4,"id":"5a890539_e42417c8","line":86,"updated":"2014-11-20 12:37:05.000000000","message":"To me, this points to us really wanting to use an image property to enable this, like hw_virtio_net_mutliqueue\n\nPossibly, we want a per flavor and/or per compute config that says whether we should respect the image property, but thats probably not needed for the first version.\n\nI don\u0027t like exposing this in the API.","commit_id":"f425f7baba83b8eabb77e52ad1a89253d4a0733b"},{"author":{"_account_id":13869,"name":"Maxime Leroy","email":"maxime.leroy@6wind.com","username":"mleroy"},"change_message_id":"bcd80e974a6595261893167d97bd08be47c22e55","unresolved":false,"context_lines":[{"line_number":83,"context_line":"    nova boot -nic -port_id -port_queues\u003d#"},{"line_number":84,"context_line":""},{"line_number":85,"context_line":"NOTE: Virtio-net multiqueue should be enabled in the guest OS manually, using"},{"line_number":86,"context_line":"ethtool."},{"line_number":87,"context_line":""},{"line_number":88,"context_line":"Alternatives"},{"line_number":89,"context_line":"------------"}],"source_content_type":"text/x-rst","patch_set":4,"id":"5a890539_a154f8fe","line":86,"in_reply_to":"5a890539_6c2b3ceb","updated":"2014-12-05 13:47:10.000000000","message":"I agree with Daniel Berrange having some NICs with a different number of queues make sense for performance requirements.\n\nHaving one NIC with one queue for management and many other NICs with several queues for performance needs, it\u0027s a common usecase when we need to run a DPDK application (www.dpdk.org) inside a guest VM.\n\nIn fact, there is one more common need to give specific parameters to a NIC like for example: enable/disable virtual HW offload (checksum, TSO, GRO) (See https://libvirt.org/formatdomain.html#elementsDriverBackendOptions)\n\nHaving an API to specify parameters to a specific port make a lot of sense.","commit_id":"f425f7baba83b8eabb77e52ad1a89253d4a0733b"},{"author":{"_account_id":782,"name":"John Garbutt","email":"john@johngarbutt.com","username":"johngarbutt"},"change_message_id":"7b7d5ae0df27df15247275bc7c6e385d65ac7b8b","unresolved":false,"context_lines":[{"line_number":83,"context_line":"    nova boot -nic -port_id -port_queues\u003d#"},{"line_number":84,"context_line":""},{"line_number":85,"context_line":"NOTE: Virtio-net multiqueue should be enabled in the guest OS manually, using"},{"line_number":86,"context_line":"ethtool."},{"line_number":87,"context_line":""},{"line_number":88,"context_line":"Alternatives"},{"line_number":89,"context_line":"------------"}],"source_content_type":"text/x-rst","patch_set":4,"id":"5a890539_6c2b3ceb","line":86,"in_reply_to":"5a890539_98bd74da","updated":"2014-12-03 17:43:55.000000000","message":"Thats a good use case, the different nics with different perf requirements.\n\nI was thinking can we fix this to just the flavor and image, but you are right, with a finite resource, and it being related to each nic, you need that control, probably.","commit_id":"f425f7baba83b8eabb77e52ad1a89253d4a0733b"},{"author":{"_account_id":1779,"name":"Daniel Berrange","email":"berrange@redhat.com","username":"berrange"},"change_message_id":"e3ed90dea82ca15e984b90e563691d74205b7548","unresolved":false,"context_lines":[{"line_number":83,"context_line":"    nova boot -nic -port_id -port_queues\u003d#"},{"line_number":84,"context_line":""},{"line_number":85,"context_line":"NOTE: Virtio-net multiqueue should be enabled in the guest OS manually, using"},{"line_number":86,"context_line":"ethtool."},{"line_number":87,"context_line":""},{"line_number":88,"context_line":"Alternatives"},{"line_number":89,"context_line":"------------"}],"source_content_type":"text/x-rst","patch_set":4,"id":"5a890539_98bd74da","line":86,"in_reply_to":"5a890539_e42417c8","updated":"2014-11-20 14:01:56.000000000","message":"While it would be nice if we could set a value for number of queues globally in the image or flavour this is just not practical.  The root problem is that we have a finite number of MSI vectors available in the guest virtual machine. These are consumed by NICs, disks, assigned PCI devices and more. In guests with large numbers of NICs and large numbers of vCPUs this is going to be a significant limiting factor. We need the ability to selectively decide which guest NICs make use of this multi queue feature, in order to avoid hitting the limit on MSIs, while still maximising use of those NICs. \n\nFor example, one NIC might be used solely for administrative traffic and so not need multiqueue enabled, while a bunch of other NICs might be used for high throughput traffic and want #queues \u003e 0. We don\u0027t want to force all NICs to have the same queue count.\n\nI\u0027m not a huge fan of exposing it in the API, but I don\u0027t see a viable alternative to control this in a fine grained manner per NIC.","commit_id":"f425f7baba83b8eabb77e52ad1a89253d4a0733b"},{"author":{"_account_id":782,"name":"John Garbutt","email":"john@johngarbutt.com","username":"johngarbutt"},"change_message_id":"bf9967785b8c127fbd11e9daf8ce01ffd308d790","unresolved":false,"context_lines":[{"line_number":144,"context_line":""},{"line_number":145,"context_line":"* Update the vif config file to set the correct number of VCPUs to"},{"line_number":146,"context_line":"  virtio-net devices"},{"line_number":147,"context_line":"* Update the NetworkRequest object to add number of queues per port"},{"line_number":148,"context_line":""},{"line_number":149,"context_line":"Dependencies"},{"line_number":150,"context_line":"\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d"}],"source_content_type":"text/x-rst","patch_set":4,"id":"5a890539_e40b3729","line":147,"updated":"2014-11-20 12:37:05.000000000","message":"I think an image property is a better fit here.\n\nWhy should the end user have to remember when and where that can enable this feature? The images should just boot with the best performance they can offer, surely?","commit_id":"f425f7baba83b8eabb77e52ad1a89253d4a0733b"},{"author":{"_account_id":6772,"name":"Stephen Gordon","email":"sgordon@redhat.com","username":"sgordon"},"change_message_id":"70ce2e370d75cad25ecf20985921876ef3f4eb1b","unresolved":false,"context_lines":[{"line_number":8,"context_line":"libvirt: virtio-net multiqueue"},{"line_number":9,"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":10,"context_line":""},{"line_number":11,"context_line":"https://blueprints.launchpad.net/nova/+spec/virtio-net-multiqueue"},{"line_number":12,"context_line":""},{"line_number":13,"context_line":"This blueprint aims to enhance the libvirt driver to enable the virtio-net"},{"line_number":14,"context_line":"multiqueue. This will allow the guest instances to increase the total network"}],"source_content_type":"text/x-rst","patch_set":5,"id":"9a80dd14_181af66e","line":11,"updated":"2015-03-13 19:27:49.000000000","message":"https://blueprints.launchpad.net/nova/+spec/libvirt-virtio-net-multiqueue","commit_id":"ace6873492778897feb01c3979ed63b5a397df6c"},{"author":{"_account_id":8802,"name":"Vladik Romanovsky","email":"vromanso@redhat.com","username":"vladikr"},"change_message_id":"213a2bae2d03373a67596e2a51bfffd6e1889dd1","unresolved":false,"context_lines":[{"line_number":8,"context_line":"libvirt: virtio-net multiqueue"},{"line_number":9,"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":10,"context_line":""},{"line_number":11,"context_line":"https://blueprints.launchpad.net/nova/+spec/virtio-net-multiqueue"},{"line_number":12,"context_line":""},{"line_number":13,"context_line":"This blueprint aims to enhance the libvirt driver to enable the virtio-net"},{"line_number":14,"context_line":"multiqueue. This will allow the guest instances to increase the total network"}],"source_content_type":"text/x-rst","patch_set":5,"id":"9a80dd14_9c7dc36f","line":11,"in_reply_to":"9a80dd14_181af66e","updated":"2015-03-16 14:38:26.000000000","message":"Done","commit_id":"ace6873492778897feb01c3979ed63b5a397df6c"},{"author":{"_account_id":3217,"name":"Ian Wells","username":"ijw-ubuntu"},"change_message_id":"71053246eff9c322a533817dfadc5c91fefd0daa","unresolved":false,"context_lines":[{"line_number":76,"context_line":"Proposed change"},{"line_number":77,"context_line":"\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d"},{"line_number":78,"context_line":""},{"line_number":79,"context_line":"In order to address the problem, an administrator should have the control"},{"line_number":80,"context_line":"over the paralleled packet processing by disabling the multiqueue support"},{"line_number":81,"context_line":"for certain workloads, where this feature may lead to a"},{"line_number":82,"context_line":"performance degradation."}],"source_content_type":"text/x-rst","patch_set":5,"id":"3a961159_018c1cd9","line":79,"updated":"2014-12-14 12:52:29.000000000","message":"Why an administrator?  If the constraint is on the number of channels per guest (per sgordon), then it seems to me like this is neither an expensive nor a (system level) scarce resource and should be an option that the user can enable, much like they can choose the NIC type at the moment.  If that\u0027s the case it makes more sense as an image property.","commit_id":"ace6873492778897feb01c3979ed63b5a397df6c"},{"author":{"_account_id":6772,"name":"Stephen Gordon","email":"sgordon@redhat.com","username":"sgordon"},"change_message_id":"d1e9c9d4b25e8991da536cde603864641362bb24","unresolved":false,"context_lines":[{"line_number":76,"context_line":"Proposed change"},{"line_number":77,"context_line":"\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d"},{"line_number":78,"context_line":""},{"line_number":79,"context_line":"In order to address the problem, an administrator should have the control"},{"line_number":80,"context_line":"over the paralleled packet processing by disabling the multiqueue support"},{"line_number":81,"context_line":"for certain workloads, where this feature may lead to a"},{"line_number":82,"context_line":"performance degradation."}],"source_content_type":"text/x-rst","patch_set":5,"id":"9a80dd14_d8c4de93","line":79,"in_reply_to":"1a930d6b_4f1f765b","updated":"2015-03-13 19:33:04.000000000","message":"+1, agree with Ian here - this should be available to any user. I would expect like other \"knobs\" in this space to be able to do from flavor extra specs or image properties.","commit_id":"ace6873492778897feb01c3979ed63b5a397df6c"},{"author":{"_account_id":8802,"name":"Vladik Romanovsky","email":"vromanso@redhat.com","username":"vladikr"},"change_message_id":"7e29e57e22895fd2276b78c149ab98d99a406b80","unresolved":false,"context_lines":[{"line_number":76,"context_line":"Proposed change"},{"line_number":77,"context_line":"\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d"},{"line_number":78,"context_line":""},{"line_number":79,"context_line":"In order to address the problem, an administrator should have the control"},{"line_number":80,"context_line":"over the paralleled packet processing by disabling the multiqueue support"},{"line_number":81,"context_line":"for certain workloads, where this feature may lead to a"},{"line_number":82,"context_line":"performance degradation."}],"source_content_type":"text/x-rst","patch_set":5,"id":"3a961159_2ca573b5","line":79,"in_reply_to":"3a961159_018c1cd9","updated":"2014-12-15 17:01:14.000000000","message":"Right, this should be enabled by default.\nHowever, I still think the cloud admin should have a control over this feature, as a wrong use of this may lead to performance degradation.","commit_id":"ace6873492778897feb01c3979ed63b5a397df6c"},{"author":{"_account_id":3217,"name":"Ian Wells","username":"ijw-ubuntu"},"change_message_id":"dc505cbb9587e9373d0354b26b380981e9fbc2eb","unresolved":false,"context_lines":[{"line_number":76,"context_line":"Proposed change"},{"line_number":77,"context_line":"\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d"},{"line_number":78,"context_line":""},{"line_number":79,"context_line":"In order to address the problem, an administrator should have the control"},{"line_number":80,"context_line":"over the paralleled packet processing by disabling the multiqueue support"},{"line_number":81,"context_line":"for certain workloads, where this feature may lead to a"},{"line_number":82,"context_line":"performance degradation."}],"source_content_type":"text/x-rst","patch_set":5,"id":"1a930d6b_4f1f765b","line":79,"in_reply_to":"3a961159_2ca573b5","updated":"2015-01-26 03:01:01.000000000","message":"I think you\u0027d have to clarify how this would degrade performance for users other than the one of the VM for which is was set.  If it degrades performance only for the VM on which it is set, then that\u0027s an option that the user or image provider can choose for themselves.  Only if it\u0027s system level degradation does it need to be administrative - otherwise you make it impossible for an app designer to make the right choise for their own situtation.","commit_id":"ace6873492778897feb01c3979ed63b5a397df6c"},{"author":{"_account_id":8802,"name":"Vladik Romanovsky","email":"vromanso@redhat.com","username":"vladikr"},"change_message_id":"213a2bae2d03373a67596e2a51bfffd6e1889dd1","unresolved":false,"context_lines":[{"line_number":76,"context_line":"Proposed change"},{"line_number":77,"context_line":"\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d"},{"line_number":78,"context_line":""},{"line_number":79,"context_line":"In order to address the problem, an administrator should have the control"},{"line_number":80,"context_line":"over the paralleled packet processing by disabling the multiqueue support"},{"line_number":81,"context_line":"for certain workloads, where this feature may lead to a"},{"line_number":82,"context_line":"performance degradation."}],"source_content_type":"text/x-rst","patch_set":5,"id":"9a80dd14_2713e401","line":79,"in_reply_to":"9a80dd14_d8c4de93","updated":"2015-03-16 14:38:26.000000000","message":"Flavor extra specs are only accessible by the administrators..\nThe idea was that the administrators can disable the feature, in general, using the extra specs, while the user can set the desired number of queues on each vnic. Demanding queue\u003d1 is equal to disabling the feature or is an explicit value in image properties is needed?\n\n\"hw_vif_multiqueue_permitted\u003dtrue|false\"\n\nI just didn\u0027t see a reason for it.","commit_id":"ace6873492778897feb01c3979ed63b5a397df6c"},{"author":{"_account_id":8976,"name":"shihanzhang","email":"ayshihanzhang@gmail.com","username":"shihanzhang"},"change_message_id":"288f9e32e809af2f0a7d3cb793cc590ef15c0519","unresolved":false,"context_lines":[{"line_number":87,"context_line":"    hw:vif_multiqueue_permitted\u003dtrue|false  (default true)"},{"line_number":88,"context_line":""},{"line_number":89,"context_line":"Users will be able to set the number of queues as part of the network port"},{"line_number":90,"context_line":"properties."},{"line_number":91,"context_line":"The default number of queues will match the number of vCPUs, defined for the"},{"line_number":92,"context_line":"instance."},{"line_number":93,"context_line":""}],"source_content_type":"text/x-rst","patch_set":5,"id":"3a961159_74082780","line":90,"updated":"2015-01-19 01:49:33.000000000","message":"what do you mean? set the \u0027queues\u0027 as a neutron port property? if so, what property?","commit_id":"ace6873492778897feb01c3979ed63b5a397df6c"},{"author":{"_account_id":8802,"name":"Vladik Romanovsky","email":"vromanso@redhat.com","username":"vladikr"},"change_message_id":"213a2bae2d03373a67596e2a51bfffd6e1889dd1","unresolved":false,"context_lines":[{"line_number":87,"context_line":"    hw:vif_multiqueue_permitted\u003dtrue|false  (default true)"},{"line_number":88,"context_line":""},{"line_number":89,"context_line":"Users will be able to set the number of queues as part of the network port"},{"line_number":90,"context_line":"properties."},{"line_number":91,"context_line":"The default number of queues will match the number of vCPUs, defined for the"},{"line_number":92,"context_line":"instance."},{"line_number":93,"context_line":""}],"source_content_type":"text/x-rst","patch_set":5,"id":"9a80dd14_47061887","line":90,"in_reply_to":"3a961159_74082780","updated":"2015-03-16 14:38:26.000000000","message":"As shown below in the example,  the -nic part creates a \nNetworkRequest object in nova, that is being used to create the relevant VIFs, while creating an instance. Queues property will be part of this object. Not in neutron itself.","commit_id":"ace6873492778897feb01c3979ed63b5a397df6c"},{"author":{"_account_id":6772,"name":"Stephen Gordon","email":"sgordon@redhat.com","username":"sgordon"},"change_message_id":"22ac48d714edd04c72e070a0635c55c72a288bf0","unresolved":false,"context_lines":[{"line_number":89,"context_line":"Users will be able to set the number of queues as part of the network port"},{"line_number":90,"context_line":"properties."},{"line_number":91,"context_line":"The default number of queues will match the number of vCPUs, defined for the"},{"line_number":92,"context_line":"instance."},{"line_number":93,"context_line":""},{"line_number":94,"context_line":"    nova boot -nic net_id\u003d\u003cid\u003e,queues\u003d#"},{"line_number":95,"context_line":"    nova interface-attach --net-id\u003d\u003cid\u003e --queues\u003d#"}],"source_content_type":"text/x-rst","patch_set":5,"id":"9a80dd14_184bd6b1","line":92,"updated":"2015-03-13 19:42:40.000000000","message":"Just to clarify, this is 1 per vCPU per NIC correct?","commit_id":"ace6873492778897feb01c3979ed63b5a397df6c"},{"author":{"_account_id":8802,"name":"Vladik Romanovsky","email":"vromanso@redhat.com","username":"vladikr"},"change_message_id":"213a2bae2d03373a67596e2a51bfffd6e1889dd1","unresolved":false,"context_lines":[{"line_number":89,"context_line":"Users will be able to set the number of queues as part of the network port"},{"line_number":90,"context_line":"properties."},{"line_number":91,"context_line":"The default number of queues will match the number of vCPUs, defined for the"},{"line_number":92,"context_line":"instance."},{"line_number":93,"context_line":""},{"line_number":94,"context_line":"    nova boot -nic net_id\u003d\u003cid\u003e,queues\u003d#"},{"line_number":95,"context_line":"    nova interface-attach --net-id\u003d\u003cid\u003e --queues\u003d#"}],"source_content_type":"text/x-rst","patch_set":5,"id":"9a80dd14_7cc80fa8","line":92,"in_reply_to":"9a80dd14_184bd6b1","updated":"2015-03-16 14:38:26.000000000","message":"Yes.","commit_id":"ace6873492778897feb01c3979ed63b5a397df6c"}],"specs/liberty/approved/libvirt-virtiomq.rst":[{"author":{"_account_id":13869,"name":"Maxime Leroy","email":"maxime.leroy@6wind.com","username":"mleroy"},"change_message_id":"d21c4e38dccefc099757195cb37f6abfb5019bd8","unresolved":false,"context_lines":[{"line_number":12,"context_line":""},{"line_number":13,"context_line":"This blueprint aims to enhance the libvirt driver to enable the virtio-net"},{"line_number":14,"context_line":"multiqueue. This will allow the guest instances to increase the total network"},{"line_number":15,"context_line":"throughput."},{"line_number":16,"context_line":""},{"line_number":17,"context_line":"Problem description"},{"line_number":18,"context_line":"\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d"}],"source_content_type":"text/x-rst","patch_set":8,"id":"7aaa499b_3a791530","line":15,"updated":"2015-04-14 16:25:38.000000000","message":"Based on this blueprint, we can implement a similar feature for vhost-user,\nwhen qemu (http://lists.gnu.org/archive/html/qemu-devel/2015-01/msg03228.html) and libvirt will support it.","commit_id":"5a2535fdf3d90dff8ad4feb07779d2ea2a3505e2"},{"author":{"_account_id":16142,"name":"Marc E. Fiuczynski","email":"marc@fiuczynski.com"},"change_message_id":"914ef20e5a9d9742e25e25d6e5a11dc40539eeb6","unresolved":false,"context_lines":[{"line_number":25,"context_line":""},{"line_number":26,"context_line":"Multiqueue virtio-net provides the greatest performance benefit when:"},{"line_number":27,"context_line":""},{"line_number":28,"context_line":"- Traffic packets are relatively large."},{"line_number":29,"context_line":"- The guest is active on many connections at the same time, with traffic"},{"line_number":30,"context_line":"  running between guests, guest to host, or guest to an external system."},{"line_number":31,"context_line":"- The number of queues is equal to the number of vCPUs. This is because"}],"source_content_type":"text/x-rst","patch_set":8,"id":"7ad32947_a8abd588","line":28,"updated":"2015-05-01 19:45:44.000000000","message":"It is actually very beneficial in ferrying many small TCP ack packets back to the guest for workloads that involve sending high volume traffic to clients that do not support TCP GRO (i.e., most consumer devices such as mobile, smart TVs, etc).","commit_id":"5a2535fdf3d90dff8ad4feb07779d2ea2a3505e2"},{"author":{"_account_id":13869,"name":"Maxime Leroy","email":"maxime.leroy@6wind.com","username":"mleroy"},"change_message_id":"d21c4e38dccefc099757195cb37f6abfb5019bd8","unresolved":false,"context_lines":[{"line_number":40,"context_line":"  Defining an instance with multiple virtio NICs and vCPUs might lead to a"},{"line_number":41,"context_line":"  possibility of hitting the guest MSI limit."},{"line_number":42,"context_line":"- virtio-net multiqueue works well for incoming traffic, but can"},{"line_number":43,"context_line":"  occasionally cause a performance degradation, for outgoing traffic."},{"line_number":44,"context_line":"- Enabling virtio-net multiqueue increases the total network throughput,"},{"line_number":45,"context_line":"  but in parallel it also increases the CPU consumption."},{"line_number":46,"context_line":"- Enabling virtio-net multiqueue in the host QEMU config, does not enable"}],"source_content_type":"text/x-rst","patch_set":8,"id":"7aaa499b_5a974148","line":43,"updated":"2015-04-14 16:25:38.000000000","message":"any reference, please ?","commit_id":"5a2535fdf3d90dff8ad4feb07779d2ea2a3505e2"},{"author":{"_account_id":8802,"name":"Vladik Romanovsky","email":"vromanso@redhat.com","username":"vladikr"},"change_message_id":"377a9c764e4949d5d37905b0c3053906e20debde","unresolved":false,"context_lines":[{"line_number":40,"context_line":"  Defining an instance with multiple virtio NICs and vCPUs might lead to a"},{"line_number":41,"context_line":"  possibility of hitting the guest MSI limit."},{"line_number":42,"context_line":"- virtio-net multiqueue works well for incoming traffic, but can"},{"line_number":43,"context_line":"  occasionally cause a performance degradation, for outgoing traffic."},{"line_number":44,"context_line":"- Enabling virtio-net multiqueue increases the total network throughput,"},{"line_number":45,"context_line":"  but in parallel it also increases the CPU consumption."},{"line_number":46,"context_line":"- Enabling virtio-net multiqueue in the host QEMU config, does not enable"}],"source_content_type":"text/x-rst","patch_set":8,"id":"5aa745d1_a348be88","line":43,"in_reply_to":"7aaa499b_5a974148","updated":"2015-04-15 19:46:11.000000000","message":"I was mostly referring to a transmission of small packets that causes a regression. This is mentioned in the link that I have referenced at the bottom of the spec and also you can find it here:\nhttp://lwn.net/Articles/505388","commit_id":"5a2535fdf3d90dff8ad4feb07779d2ea2a3505e2"},{"author":{"_account_id":16142,"name":"Marc E. Fiuczynski","email":"marc@fiuczynski.com"},"change_message_id":"914ef20e5a9d9742e25e25d6e5a11dc40539eeb6","unresolved":false,"context_lines":[{"line_number":49,"context_line":"  ethtool."},{"line_number":50,"context_line":"- MSI vectors would still be consumed (wasted), if multiqueue was enabled"},{"line_number":51,"context_line":"  in the host, but has not been enabled in the guest OS by the administrator."},{"line_number":52,"context_line":"- In case the number of vNICs in a guest instance is proportional to the"},{"line_number":53,"context_line":"  number of vCPUs, enabling the multiqueue feature is less important."},{"line_number":54,"context_line":"- Each virtio-net queue consumes 64 KB of kernel memory for the vhost driver."},{"line_number":55,"context_line":""}],"source_content_type":"text/x-rst","patch_set":8,"id":"7ad32947_a8cab59f","line":52,"updated":"2015-05-01 19:45:44.000000000","message":"This comment assumes traffic is equally distributed across the vNICs. If it is not, then mqueue helps with small packets on the RX side.","commit_id":"5a2535fdf3d90dff8ad4feb07779d2ea2a3505e2"},{"author":{"_account_id":16142,"name":"Marc E. Fiuczynski","email":"marc@fiuczynski.com"},"change_message_id":"914ef20e5a9d9742e25e25d6e5a11dc40539eeb6","unresolved":false,"context_lines":[{"line_number":51,"context_line":"  in the host, but has not been enabled in the guest OS by the administrator."},{"line_number":52,"context_line":"- In case the number of vNICs in a guest instance is proportional to the"},{"line_number":53,"context_line":"  number of vCPUs, enabling the multiqueue feature is less important."},{"line_number":54,"context_line":"- Each virtio-net queue consumes 64 KB of kernel memory for the vhost driver."},{"line_number":55,"context_line":""},{"line_number":56,"context_line":""},{"line_number":57,"context_line":"Use Cases"}],"source_content_type":"text/x-rst","patch_set":8,"id":"7ad32947_888819e9","line":54,"updated":"2015-05-01 19:45:44.000000000","message":"And the vhost driver also has a dedicated thread per virtio-net queue. Is that capturd by the 64KB of kernel memory you are mentioning?","commit_id":"5a2535fdf3d90dff8ad4feb07779d2ea2a3505e2"},{"author":{"_account_id":782,"name":"John Garbutt","email":"john@johngarbutt.com","username":"johngarbutt"},"change_message_id":"8441c0b49ad536711edff679fad542452f20d97d","unresolved":false,"context_lines":[{"line_number":87,"context_line":"    hw:vif_multiqueue_permitted\u003dtrue|false  (default true)"},{"line_number":88,"context_line":""},{"line_number":89,"context_line":"Users will be able to set the number of queues as part of the network port"},{"line_number":90,"context_line":"properties."},{"line_number":91,"context_line":"The default number of queues will match the number of vCPUs, defined for the"},{"line_number":92,"context_line":"instance."},{"line_number":93,"context_line":""}],"source_content_type":"text/x-rst","patch_set":8,"id":"7ad32947_d3826bc8","line":90,"updated":"2015-04-27 17:01:54.000000000","message":"This doesn\u0027t feel like something the user should be picking?\n\nWhy can\u0027t this just be set at the host level? or some kind of scaling factor inside the flavor?","commit_id":"5a2535fdf3d90dff8ad4feb07779d2ea2a3505e2"},{"author":{"_account_id":1779,"name":"Daniel Berrange","email":"berrange@redhat.com","username":"berrange"},"change_message_id":"f7eb189eda235c0be546199a808acb0e0bf99dd3","unresolved":false,"context_lines":[{"line_number":87,"context_line":"    hw:vif_multiqueue_permitted\u003dtrue|false  (default true)"},{"line_number":88,"context_line":""},{"line_number":89,"context_line":"Users will be able to set the number of queues as part of the network port"},{"line_number":90,"context_line":"properties."},{"line_number":91,"context_line":"The default number of queues will match the number of vCPUs, defined for the"},{"line_number":92,"context_line":"instance."},{"line_number":93,"context_line":""}],"source_content_type":"text/x-rst","patch_set":8,"id":"fa32b979_15e89b7b","line":90,"in_reply_to":"5afe65bd_b3bbffe4","updated":"2015-06-22 14:07:06.000000000","message":"Perhaps if we wish to move this forward, we should consider doing the work in two phases. It seems people are generally happy at the concept of allowing multiqueue to be enabled/disabled, but our debate is around how we customize the number of queues per VIF.\n\nWe could start with a simple impl, where we just have a hardcoded policy for the number of queues we assign, when multiqueue is enabled. As mentioned before, this would not be satisfactory for all use cases, but it might be sufficient for a reasonable number of use cases. We could then spend more time considering how best to express the greater configurability.","commit_id":"5a2535fdf3d90dff8ad4feb07779d2ea2a3505e2"},{"author":{"_account_id":8802,"name":"Vladik Romanovsky","email":"vromanso@redhat.com","username":"vladikr"},"change_message_id":"61f66580cf59508e9cf6bd2dd72eb01a46461e7f","unresolved":false,"context_lines":[{"line_number":87,"context_line":"    hw:vif_multiqueue_permitted\u003dtrue|false  (default true)"},{"line_number":88,"context_line":""},{"line_number":89,"context_line":"Users will be able to set the number of queues as part of the network port"},{"line_number":90,"context_line":"properties."},{"line_number":91,"context_line":"The default number of queues will match the number of vCPUs, defined for the"},{"line_number":92,"context_line":"instance."},{"line_number":93,"context_line":""}],"source_content_type":"text/x-rst","patch_set":8,"id":"7ad32947_ec98e46c","line":90,"in_reply_to":"7ad32947_d3826bc8","updated":"2015-04-27 18:18:15.000000000","message":"In case this feature is permitted in the flavor and the number of queues isn\u0027t specified, as part of the CLI commands, the number of queues will be set to the number of vCPU in the guest, so the user doesn\u0027t need to know anything, unless he wants to.\n\nOne of the reason why the user should have the ability to control the number of queues per vNIC is because some networks are served for different purposes and some doesn\u0027t require high throughput.\n\nI think this has been already discussed and there is a great explanation by Daniel Berrange here, line 83 and 86 (unless I\u0027m missing something..):\nhttps://review.openstack.org/#/c/128825/4/specs/kilo/approved/libvirt-virtiomq.rst","commit_id":"5a2535fdf3d90dff8ad4feb07779d2ea2a3505e2"},{"author":{"_account_id":4393,"name":"Dan Smith","email":"dms@danplanet.com","username":"danms"},"change_message_id":"f888aedda340fbcb524be7fddb0b15ba2728a66a","unresolved":false,"context_lines":[{"line_number":87,"context_line":"    hw:vif_multiqueue_permitted\u003dtrue|false  (default true)"},{"line_number":88,"context_line":""},{"line_number":89,"context_line":"Users will be able to set the number of queues as part of the network port"},{"line_number":90,"context_line":"properties."},{"line_number":91,"context_line":"The default number of queues will match the number of vCPUs, defined for the"},{"line_number":92,"context_line":"instance."},{"line_number":93,"context_line":""}],"source_content_type":"text/x-rst","patch_set":8,"id":"9a0c5dc1_984a9d82","line":90,"in_reply_to":"7ad32947_ec98e46c","updated":"2015-05-12 18:53:51.000000000","message":"Yeah, it seems weird to let the user specify this without restriction. You say there is a 64kb cost in the guest kernel, per queue.. What about in the host? If I boot a couple guests with a bunch of nics that are all many queues, can I unnecessarily eat up a bunch of host memory? I know, it\u0027d have to be a lot to matter, but if everyone always boots with a bunch to try to improve performance, it can add up.\n\nMaybe the property should be \"allowed_number_of_queues\", where the default is 1 (current behavior) and the user can request up to that many queues per nic? That way the admin can set this to 4 or something per flavor and limit your ability to consume resources. Also, seems like the number of queues should be limited to N per vNIC, where N is the number of vcpus in the guest, regardless of any other limit, right?\n\nJohn, he did say that he wants the user to be able to specify the number per NIC because there are a limited number of slots per VM, and they can control their rates by allocating more or less channels to a given NIC. While that seems like a relatively annoying limitation of virtio (or maybe it\u0027s an intel thing), not letting them control them at all does preempt their ability to do that.\n\nIt would be nice to achieve the more automaticness goal by abstracting the \"queues\" concept and just make it a weight thing where you allocate 100% of your bandwidth chips across your NICs, which would apply to other hypervisors that probably have similar issues, but that don\u0027t exactly match the queues concept, right?","commit_id":"5a2535fdf3d90dff8ad4feb07779d2ea2a3505e2"},{"author":{"_account_id":8802,"name":"Vladik Romanovsky","email":"vromanso@redhat.com","username":"vladikr"},"change_message_id":"b0c052627f1f28bda5dcd965f8ffd5fc2f7ad768","unresolved":false,"context_lines":[{"line_number":87,"context_line":"    hw:vif_multiqueue_permitted\u003dtrue|false  (default true)"},{"line_number":88,"context_line":""},{"line_number":89,"context_line":"Users will be able to set the number of queues as part of the network port"},{"line_number":90,"context_line":"properties."},{"line_number":91,"context_line":"The default number of queues will match the number of vCPUs, defined for the"},{"line_number":92,"context_line":"instance."},{"line_number":93,"context_line":""}],"source_content_type":"text/x-rst","patch_set":8,"id":"5afe65bd_b3bbffe4","line":90,"in_reply_to":"9a0c5dc1_984a9d82","updated":"2015-06-03 18:46:56.000000000","message":"The memory used should be in the context of the VM memory and this is all about the guest resources, not the host\u0027s.\n\nI don\u0027t see a reason for the administrator to limit the number queues. The goal is to control whether the feature is enabled or disabled per flavor.\n\nIt\u0027s really up to the user to set the number of the desired queues for each nic, therefor, I think it\u0027s better to set it for each NIC via the api.\n(As I mentioned earlier, Daniel has provided a very good explanation:\n  https://review.openstack.org/#/c/128825/4/specs/kilo/approved/libvirt-virtiomq.rst)\n\nThe users will use that option to reduce the number or disable the queues per nic, not setting more, but limiting it is a good idea.\n\n\"queues\" is a general option, even for a real hardware NICs, various hypervisors can implement it. I don\u0027t see it as Libvirt specific.\n\nAs an alternative, I can suggest renaming this option to \"throughput\"\u003dhigh|medium|low that would assign 100%|50%|1 of vCPUs\n(or have a config for that)\nHowever, I still think that letting the users set the number of a desired queues is better, as they know their system best, while it doesn\u0027t affect the hypervisor resources.\n\nIf people are in general opposing to the idea of introducing this option, as a first step to get this spec from this impasse, I will change the spec to unconditionally enable all queues for each NIC, if the flavor allows it.","commit_id":"5a2535fdf3d90dff8ad4feb07779d2ea2a3505e2"},{"author":{"_account_id":782,"name":"John Garbutt","email":"john@johngarbutt.com","username":"johngarbutt"},"change_message_id":"8441c0b49ad536711edff679fad542452f20d97d","unresolved":false,"context_lines":[{"line_number":95,"context_line":"    nova interface-attach --net-id\u003d\u003cid\u003e --queues\u003d#"},{"line_number":96,"context_line":""},{"line_number":97,"context_line":"NOTE: Virtio-net multiqueue should be enabled in the guest OS manually, using"},{"line_number":98,"context_line":"ethtool."},{"line_number":99,"context_line":""},{"line_number":100,"context_line":"Alternatives"},{"line_number":101,"context_line":"------------"}],"source_content_type":"text/x-rst","patch_set":8,"id":"7ad32947_73515762","line":98,"updated":"2015-04-27 17:01:54.000000000","message":"This suggests we need an image property to say an image is able to use this feature? And fail builds that don\u0027t need that maybe?\n\nCan we please link to a reference that describes that process?","commit_id":"5a2535fdf3d90dff8ad4feb07779d2ea2a3505e2"},{"author":{"_account_id":8802,"name":"Vladik Romanovsky","email":"vromanso@redhat.com","username":"vladikr"},"change_message_id":"61f66580cf59508e9cf6bd2dd72eb01a46461e7f","unresolved":false,"context_lines":[{"line_number":95,"context_line":"    nova interface-attach --net-id\u003d\u003cid\u003e --queues\u003d#"},{"line_number":96,"context_line":""},{"line_number":97,"context_line":"NOTE: Virtio-net multiqueue should be enabled in the guest OS manually, using"},{"line_number":98,"context_line":"ethtool."},{"line_number":99,"context_line":""},{"line_number":100,"context_line":"Alternatives"},{"line_number":101,"context_line":"------------"}],"source_content_type":"text/x-rst","patch_set":8,"id":"7ad32947_c70d0139","line":98,"in_reply_to":"7ad32947_73515762","updated":"2015-04-27 18:18:15.000000000","message":"There is no need for an image property here..\nThe administrator of a guest, that has been built with this option enabled in OpenStack, can enable the multiqueue setting, by issuing ethtool command:\nethtool -L ethX combined \u003cnum_of_queues\u003e\n \nThe process is described in the referenced document under the title:\n\"Enable MQ feature\"","commit_id":"5a2535fdf3d90dff8ad4feb07779d2ea2a3505e2"},{"author":{"_account_id":4393,"name":"Dan Smith","email":"dms@danplanet.com","username":"danms"},"change_message_id":"f888aedda340fbcb524be7fddb0b15ba2728a66a","unresolved":false,"context_lines":[{"line_number":95,"context_line":"    nova interface-attach --net-id\u003d\u003cid\u003e --queues\u003d#"},{"line_number":96,"context_line":""},{"line_number":97,"context_line":"NOTE: Virtio-net multiqueue should be enabled in the guest OS manually, using"},{"line_number":98,"context_line":"ethtool."},{"line_number":99,"context_line":""},{"line_number":100,"context_line":"Alternatives"},{"line_number":101,"context_line":"------------"}],"source_content_type":"text/x-rst","patch_set":8,"id":"9a0c5dc1_b81d61bf","line":98,"in_reply_to":"7ad32947_c70d0139","updated":"2015-05-12 18:53:51.000000000","message":"Does this mean that if we enable MQ and then boot an old guest that doesn\u0027t support it, that the guest will just use the first queue and be happy?","commit_id":"5a2535fdf3d90dff8ad4feb07779d2ea2a3505e2"},{"author":{"_account_id":8802,"name":"Vladik Romanovsky","email":"vromanso@redhat.com","username":"vladikr"},"change_message_id":"b0c052627f1f28bda5dcd965f8ffd5fc2f7ad768","unresolved":false,"context_lines":[{"line_number":95,"context_line":"    nova interface-attach --net-id\u003d\u003cid\u003e --queues\u003d#"},{"line_number":96,"context_line":""},{"line_number":97,"context_line":"NOTE: Virtio-net multiqueue should be enabled in the guest OS manually, using"},{"line_number":98,"context_line":"ethtool."},{"line_number":99,"context_line":""},{"line_number":100,"context_line":"Alternatives"},{"line_number":101,"context_line":"------------"}],"source_content_type":"text/x-rst","patch_set":8,"id":"5afe65bd_333cafc7","line":98,"in_reply_to":"9a0c5dc1_b81d61bf","updated":"2015-06-03 18:46:56.000000000","message":"Yes. MQ is not getting enabled by default. It has to be manually enabled in the the guest OS for each NIC. However, the reserved MSIs will be wasted..","commit_id":"5a2535fdf3d90dff8ad4feb07779d2ea2a3505e2"},{"author":{"_account_id":4393,"name":"Dan Smith","email":"dms@danplanet.com","username":"danms"},"change_message_id":"f888aedda340fbcb524be7fddb0b15ba2728a66a","unresolved":false,"context_lines":[{"line_number":100,"context_line":"Alternatives"},{"line_number":101,"context_line":"------------"},{"line_number":102,"context_line":""},{"line_number":103,"context_line":"None"},{"line_number":104,"context_line":""},{"line_number":105,"context_line":"Data model impact"},{"line_number":106,"context_line":"-----------------"}],"source_content_type":"text/x-rst","patch_set":8,"id":"9a0c5dc1_5812e563","line":103,"updated":"2015-05-12 18:53:51.000000000","message":"IMHO, there are several alternatives popping up above. If we were to go with what you have here, an alternative would be \"allow the admin to specify the number on the flavor, config, etc\"","commit_id":"5a2535fdf3d90dff8ad4feb07779d2ea2a3505e2"},{"author":{"_account_id":8802,"name":"Vladik Romanovsky","email":"vromanso@redhat.com","username":"vladikr"},"change_message_id":"b0c052627f1f28bda5dcd965f8ffd5fc2f7ad768","unresolved":false,"context_lines":[{"line_number":100,"context_line":"Alternatives"},{"line_number":101,"context_line":"------------"},{"line_number":102,"context_line":""},{"line_number":103,"context_line":"None"},{"line_number":104,"context_line":""},{"line_number":105,"context_line":"Data model impact"},{"line_number":106,"context_line":"-----------------"}],"source_content_type":"text/x-rst","patch_set":8,"id":"5afe65bd_d330bbea","line":103,"in_reply_to":"9a0c5dc1_5812e563","updated":"2015-06-03 18:46:56.000000000","message":"OK,I\u0027ll add that.","commit_id":"5a2535fdf3d90dff8ad4feb07779d2ea2a3505e2"},{"author":{"_account_id":4393,"name":"Dan Smith","email":"dms@danplanet.com","username":"danms"},"change_message_id":"f888aedda340fbcb524be7fddb0b15ba2728a66a","unresolved":false,"context_lines":[{"line_number":135,"context_line":"Other deployer impact"},{"line_number":136,"context_line":"---------------------"},{"line_number":137,"context_line":""},{"line_number":138,"context_line":"None."},{"line_number":139,"context_line":""},{"line_number":140,"context_line":"Developer impact"},{"line_number":141,"context_line":"----------------"}],"source_content_type":"text/x-rst","patch_set":8,"id":"9a0c5dc1_185fcd42","line":138,"updated":"2015-05-12 18:53:51.000000000","message":"There is some deployer impact here because you\u0027re defaulting \u003dTrue. If this is not a thing that current deployers want, then they need to set this key on all their flavors before upgrading, right?","commit_id":"5a2535fdf3d90dff8ad4feb07779d2ea2a3505e2"},{"author":{"_account_id":8802,"name":"Vladik Romanovsky","email":"vromanso@redhat.com","username":"vladikr"},"change_message_id":"b0c052627f1f28bda5dcd965f8ffd5fc2f7ad768","unresolved":false,"context_lines":[{"line_number":135,"context_line":"Other deployer impact"},{"line_number":136,"context_line":"---------------------"},{"line_number":137,"context_line":""},{"line_number":138,"context_line":"None."},{"line_number":139,"context_line":""},{"line_number":140,"context_line":"Developer impact"},{"line_number":141,"context_line":"----------------"}],"source_content_type":"text/x-rst","patch_set":8,"id":"5afe65bd_9386c363","line":138,"in_reply_to":"9a0c5dc1_185fcd42","updated":"2015-06-03 18:46:56.000000000","message":"I think most deployers will want that to be configured by default. Since the guest admin can control the enablement of the feature in the guests, it will just be less configurations for them to do, but I\u0027ll mention it here.","commit_id":"5a2535fdf3d90dff8ad4feb07779d2ea2a3505e2"},{"author":{"_account_id":7,"name":"Jay Pipes","email":"jaypipes@gmail.com","username":"jaypipes"},"change_message_id":"4f93765f3942f4b0f0b145750b8e20a8c77b015a","unresolved":false,"context_lines":[{"line_number":83,"context_line":"instance."},{"line_number":84,"context_line":""},{"line_number":85,"context_line":"NOTE: Virtio-net multiqueue should be enabled in the guest OS manually, using"},{"line_number":86,"context_line":"ethtool."},{"line_number":87,"context_line":""},{"line_number":88,"context_line":"Alternatives"},{"line_number":89,"context_line":"------------"}],"source_content_type":"text/x-rst","patch_set":11,"id":"fa32b979_4a7d071b","line":86,"updated":"2015-06-22 22:35:45.000000000","message":"You might consider putting the instructions for enablement here...","commit_id":"24441d0c039597be7a20416547312a0d66347b2a"},{"author":{"_account_id":8802,"name":"Vladik Romanovsky","email":"vromanso@redhat.com","username":"vladikr"},"change_message_id":"9d083722bf4e1cbae1c66d8fc7bff6458e004299","unresolved":false,"context_lines":[{"line_number":83,"context_line":"instance."},{"line_number":84,"context_line":""},{"line_number":85,"context_line":"NOTE: Virtio-net multiqueue should be enabled in the guest OS manually, using"},{"line_number":86,"context_line":"ethtool."},{"line_number":87,"context_line":""},{"line_number":88,"context_line":"Alternatives"},{"line_number":89,"context_line":"------------"}],"source_content_type":"text/x-rst","patch_set":11,"id":"fa32b979_b5be5aac","line":86,"in_reply_to":"fa32b979_4a7d071b","updated":"2015-06-23 13:00:11.000000000","message":"Done","commit_id":"24441d0c039597be7a20416547312a0d66347b2a"},{"author":{"_account_id":7,"name":"Jay Pipes","email":"jaypipes@gmail.com","username":"jaypipes"},"change_message_id":"4f93765f3942f4b0f0b145750b8e20a8c77b015a","unresolved":false,"context_lines":[{"line_number":150,"context_line":"feature will not get enabled in the guest NICs MSI vectors will be consumed."},{"line_number":151,"context_line":"In order to avoid the MSI vectors consumption, for those who are not"},{"line_number":152,"context_line":"interested in the feature, the feature should be disabled in the image"},{"line_number":153,"context_line":"propertes by setting \"hw_vif_multiqueue_enabled\u003dfalse\""},{"line_number":154,"context_line":""},{"line_number":155,"context_line":"Developer impact"},{"line_number":156,"context_line":"----------------"}],"source_content_type":"text/x-rst","patch_set":11,"id":"fa32b979_8ab0efde","line":153,"updated":"2015-06-22 22:35:45.000000000","message":"Why not have it disabled by default and force the cloud admin to enable it on specific images?","commit_id":"24441d0c039597be7a20416547312a0d66347b2a"},{"author":{"_account_id":8802,"name":"Vladik Romanovsky","email":"vromanso@redhat.com","username":"vladikr"},"change_message_id":"9d083722bf4e1cbae1c66d8fc7bff6458e004299","unresolved":false,"context_lines":[{"line_number":150,"context_line":"feature will not get enabled in the guest NICs MSI vectors will be consumed."},{"line_number":151,"context_line":"In order to avoid the MSI vectors consumption, for those who are not"},{"line_number":152,"context_line":"interested in the feature, the feature should be disabled in the image"},{"line_number":153,"context_line":"propertes by setting \"hw_vif_multiqueue_enabled\u003dfalse\""},{"line_number":154,"context_line":""},{"line_number":155,"context_line":"Developer impact"},{"line_number":156,"context_line":"----------------"}],"source_content_type":"text/x-rst","patch_set":11,"id":"fa32b979_55deee11","line":153,"in_reply_to":"fa32b979_8ab0efde","updated":"2015-06-23 13:00:11.000000000","message":"Done","commit_id":"24441d0c039597be7a20416547312a0d66347b2a"},{"author":{"_account_id":1779,"name":"Daniel Berrange","email":"berrange@redhat.com","username":"berrange"},"change_message_id":"b6c6237c32057adc39df85f95bfb4c293cdd30f3","unresolved":false,"context_lines":[{"line_number":150,"context_line":"feature will not get enabled in the guest NICs MSI vectors will be consumed."},{"line_number":151,"context_line":"In order to avoid the MSI vectors consumption, for those who are not"},{"line_number":152,"context_line":"interested in the feature, the feature should be disabled in the image"},{"line_number":153,"context_line":"propertes by setting \"hw_vif_multiqueue_enabled\u003dfalse\""},{"line_number":154,"context_line":""},{"line_number":155,"context_line":"Developer impact"},{"line_number":156,"context_line":"----------------"}],"source_content_type":"text/x-rst","patch_set":11,"id":"fa32b979_fe412f3a","line":153,"in_reply_to":"fa32b979_8ab0efde","updated":"2015-06-23 09:14:37.000000000","message":"Yep, the concern with enabling it by default is that we could cause a regression in ability to boot guests with large numbers of nics \u0026 vcpus, as we\u0027d hit the MSI limit immediately. The safe option is to disable it by default I think","commit_id":"24441d0c039597be7a20416547312a0d66347b2a"},{"author":{"_account_id":4393,"name":"Dan Smith","email":"dms@danplanet.com","username":"danms"},"change_message_id":"7b160dd6aec0428e667ad77552587a58ccc0a390","unresolved":false,"context_lines":[{"line_number":18,"context_line":"\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d"},{"line_number":19,"context_line":""},{"line_number":20,"context_line":"Today\u0027s high-end servers have more processors, and guests running on them often"},{"line_number":21,"context_line":"have an increasing number of vCPUs. In а single virtio-net queue, the scale of"},{"line_number":22,"context_line":"the protocol stack in a guest is restricted, as the network performance does"},{"line_number":23,"context_line":"not scale as the number of vCPUs increases. Guests cannot transmit or"},{"line_number":24,"context_line":"retrieve packets in parallel, as virtio-net has only one TX and RX queue."}],"source_content_type":"text/x-rst","patch_set":12,"id":"fa32b979_fcdcc6e5","line":21,"range":{"start_line":21,"start_character":39,"end_line":21,"end_character":40},"updated":"2015-06-23 15:12:41.000000000","message":"This looks like a unicode symbol and not an ascii lowercase \"A\". Is that just my imagination?","commit_id":"46739f5db7acbed486f5e75b1d77a0d03a027023"}],"specs/liberty/libvirt-virtiomq.rst":[{"author":{"_account_id":7730,"name":"Sahid Orentino Ferdjaoui","email":"sahid.ferdjaoui@industrialdiscipline.com","username":"sahid"},"change_message_id":"46bbbec4c5334df2501ed000ab4a9673178e8a3c","unresolved":false,"context_lines":[{"line_number":84,"context_line":"Introducing a new parameter to the flavor extra specs, for the administrator"},{"line_number":85,"context_line":"to control the virtio-net multiqueue feature."},{"line_number":86,"context_line":""},{"line_number":87,"context_line":"    hw:vif_multiqueue_permitted\u003dtrue|false  (default true)"},{"line_number":88,"context_line":""},{"line_number":89,"context_line":"Users will be able to set the number of queues as part of the network port"},{"line_number":90,"context_line":"properties."}],"source_content_type":"text/x-rst","patch_set":7,"id":"9a80dd14_d73448fd","line":87,"updated":"2015-03-17 10:33:19.000000000","message":"Can we give a possibility to the operator to control the max number of queues allowed?\n\n  - hw:vif_multiqueue_permitted\u003dtrue|false|\u003cmax-permitted\u003e\n\nWith:\n - True: with a max of the number of vCPU\n - False: not permitted\n - Int: max number of queue permitted by the operator","commit_id":"65b49f330669c11695f70ea744cd3e775fad9962"},{"author":{"_account_id":7730,"name":"Sahid Orentino Ferdjaoui","email":"sahid.ferdjaoui@industrialdiscipline.com","username":"sahid"},"change_message_id":"8ffd03eadd56a007f4d54f64db065797d0c163d3","unresolved":false,"context_lines":[{"line_number":84,"context_line":"Introducing a new parameter to the flavor extra specs, for the administrator"},{"line_number":85,"context_line":"to control the virtio-net multiqueue feature."},{"line_number":86,"context_line":""},{"line_number":87,"context_line":"    hw:vif_multiqueue_permitted\u003dtrue|false  (default true)"},{"line_number":88,"context_line":""},{"line_number":89,"context_line":"Users will be able to set the number of queues as part of the network port"},{"line_number":90,"context_line":"properties."}],"source_content_type":"text/x-rst","patch_set":7,"id":"1a6ced46_b31f886b","line":87,"in_reply_to":"1a6ced46_0f305155","updated":"2015-03-25 11:15:58.000000000","message":"Is there no impact on the host if my instance ask for 100 queues?","commit_id":"65b49f330669c11695f70ea744cd3e775fad9962"},{"author":{"_account_id":7730,"name":"Sahid Orentino Ferdjaoui","email":"sahid.ferdjaoui@industrialdiscipline.com","username":"sahid"},"change_message_id":"84ec9bc78d1aad89def0ecaf7dee5bc215a0fba3","unresolved":false,"context_lines":[{"line_number":84,"context_line":"Introducing a new parameter to the flavor extra specs, for the administrator"},{"line_number":85,"context_line":"to control the virtio-net multiqueue feature."},{"line_number":86,"context_line":""},{"line_number":87,"context_line":"    hw:vif_multiqueue_permitted\u003dtrue|false  (default true)"},{"line_number":88,"context_line":""},{"line_number":89,"context_line":"Users will be able to set the number of queues as part of the network port"},{"line_number":90,"context_line":"properties."}],"source_content_type":"text/x-rst","patch_set":7,"id":"1a6ced46_45f01dfe","line":87,"in_reply_to":"1a6ced46_a7400b35","updated":"2015-03-26 11:50:12.000000000","message":"Ok, if that mostly affecting guest and perhaps we can think about such limitation when implementing code.","commit_id":"65b49f330669c11695f70ea744cd3e775fad9962"},{"author":{"_account_id":8802,"name":"Vladik Romanovsky","email":"vromanso@redhat.com","username":"vladikr"},"change_message_id":"acb4d8b30b0ec82d201e1b51cd61f57fb6764532","unresolved":false,"context_lines":[{"line_number":84,"context_line":"Introducing a new parameter to the flavor extra specs, for the administrator"},{"line_number":85,"context_line":"to control the virtio-net multiqueue feature."},{"line_number":86,"context_line":""},{"line_number":87,"context_line":"    hw:vif_multiqueue_permitted\u003dtrue|false  (default true)"},{"line_number":88,"context_line":""},{"line_number":89,"context_line":"Users will be able to set the number of queues as part of the network port"},{"line_number":90,"context_line":"properties."}],"source_content_type":"text/x-rst","patch_set":7,"id":"1a6ced46_a7400b35","line":87,"in_reply_to":"1a6ced46_b31f886b","updated":"2015-03-25 20:50:29.000000000","message":"This feature is mostly affecting the guest.. with all the limitations listed in the problem description above.","commit_id":"65b49f330669c11695f70ea744cd3e775fad9962"},{"author":{"_account_id":8802,"name":"Vladik Romanovsky","email":"vromanso@redhat.com","username":"vladikr"},"change_message_id":"4dbd450ef9ef32c386ae800bf6ea436d5ba083c4","unresolved":false,"context_lines":[{"line_number":84,"context_line":"Introducing a new parameter to the flavor extra specs, for the administrator"},{"line_number":85,"context_line":"to control the virtio-net multiqueue feature."},{"line_number":86,"context_line":""},{"line_number":87,"context_line":"    hw:vif_multiqueue_permitted\u003dtrue|false  (default true)"},{"line_number":88,"context_line":""},{"line_number":89,"context_line":"Users will be able to set the number of queues as part of the network port"},{"line_number":90,"context_line":"properties."}],"source_content_type":"text/x-rst","patch_set":7,"id":"1a6ced46_0f305155","line":87,"in_reply_to":"9a80dd14_d73448fd","updated":"2015-03-24 17:33:19.000000000","message":"I\u0027m not sure why is it needed. The user needs to control this number on a vm based level, as oppose to the cloud admin who can only control this number on a flavor level. \nI don\u0027t see a reason to limit it on a flavor level, as this is not a host resource.\nThis option is only needed for the admin to disable the feature entirely, for whatever reason. Or am I missing anything?","commit_id":"65b49f330669c11695f70ea744cd3e775fad9962"},{"author":{"_account_id":7730,"name":"Sahid Orentino Ferdjaoui","email":"sahid.ferdjaoui@industrialdiscipline.com","username":"sahid"},"change_message_id":"46bbbec4c5334df2501ed000ab4a9673178e8a3c","unresolved":false,"context_lines":[{"line_number":92,"context_line":"instance."},{"line_number":93,"context_line":""},{"line_number":94,"context_line":"    nova boot -nic net_id\u003d\u003cid\u003e,queues\u003d#"},{"line_number":95,"context_line":"    nova interface-attach --net-id\u003d\u003cid\u003e --queues\u003d#"},{"line_number":96,"context_line":""},{"line_number":97,"context_line":"NOTE: Virtio-net multiqueue should be enabled in the guest OS manually, using"},{"line_number":98,"context_line":"ethtool."}],"source_content_type":"text/x-rst","patch_set":7,"id":"9a80dd14_526306dd","line":95,"updated":"2015-03-17 10:33:19.000000000","message":"Since the Guest have to be configured (ethtool), does it make more sense to have this option passed by the image-meta property?","commit_id":"65b49f330669c11695f70ea744cd3e775fad9962"},{"author":{"_account_id":8802,"name":"Vladik Romanovsky","email":"vromanso@redhat.com","username":"vladikr"},"change_message_id":"acb4d8b30b0ec82d201e1b51cd61f57fb6764532","unresolved":false,"context_lines":[{"line_number":92,"context_line":"instance."},{"line_number":93,"context_line":""},{"line_number":94,"context_line":"    nova boot -nic net_id\u003d\u003cid\u003e,queues\u003d#"},{"line_number":95,"context_line":"    nova interface-attach --net-id\u003d\u003cid\u003e --queues\u003d#"},{"line_number":96,"context_line":""},{"line_number":97,"context_line":"NOTE: Virtio-net multiqueue should be enabled in the guest OS manually, using"},{"line_number":98,"context_line":"ethtool."}],"source_content_type":"text/x-rst","patch_set":7,"id":"1a6ced46_a7956b88","line":95,"in_reply_to":"1a6ced46_332478c4","updated":"2015-03-25 20:50:29.000000000","message":"We will need a \"queues\" property per each vNic. The more vNics we will have the more complex the image property configuration will be..\n\nI think Daniel has a good explanation on:\nhttps://review.openstack.org/#/c/128825/4/specs/kilo/approved/libvirt-virtiomq.rst , Comment on line 86.","commit_id":"65b49f330669c11695f70ea744cd3e775fad9962"},{"author":{"_account_id":7730,"name":"Sahid Orentino Ferdjaoui","email":"sahid.ferdjaoui@industrialdiscipline.com","username":"sahid"},"change_message_id":"84ec9bc78d1aad89def0ecaf7dee5bc215a0fba3","unresolved":false,"context_lines":[{"line_number":92,"context_line":"instance."},{"line_number":93,"context_line":""},{"line_number":94,"context_line":"    nova boot -nic net_id\u003d\u003cid\u003e,queues\u003d#"},{"line_number":95,"context_line":"    nova interface-attach --net-id\u003d\u003cid\u003e --queues\u003d#"},{"line_number":96,"context_line":""},{"line_number":97,"context_line":"NOTE: Virtio-net multiqueue should be enabled in the guest OS manually, using"},{"line_number":98,"context_line":"ethtool."}],"source_content_type":"text/x-rst","patch_set":7,"id":"1a6ced46_9627b111","line":95,"in_reply_to":"1a6ced46_a7956b88","updated":"2015-03-26 11:50:12.000000000","message":"Yes... I complain on something which was already responded, sorry Vladik","commit_id":"65b49f330669c11695f70ea744cd3e775fad9962"},{"author":{"_account_id":7730,"name":"Sahid Orentino Ferdjaoui","email":"sahid.ferdjaoui@industrialdiscipline.com","username":"sahid"},"change_message_id":"8ffd03eadd56a007f4d54f64db065797d0c163d3","unresolved":false,"context_lines":[{"line_number":92,"context_line":"instance."},{"line_number":93,"context_line":""},{"line_number":94,"context_line":"    nova boot -nic net_id\u003d\u003cid\u003e,queues\u003d#"},{"line_number":95,"context_line":"    nova interface-attach --net-id\u003d\u003cid\u003e --queues\u003d#"},{"line_number":96,"context_line":""},{"line_number":97,"context_line":"NOTE: Virtio-net multiqueue should be enabled in the guest OS manually, using"},{"line_number":98,"context_line":"ethtool."}],"source_content_type":"text/x-rst","patch_set":7,"id":"1a6ced46_332478c4","line":95,"in_reply_to":"1a6ced46_afa01da0","updated":"2015-03-25 11:15:58.000000000","message":"You mean is it not possible to have an image already well configured to handle queues and then one image property (e.g \"queue\") to \"hardcore\" the number of queues needed by the image?","commit_id":"65b49f330669c11695f70ea744cd3e775fad9962"},{"author":{"_account_id":8802,"name":"Vladik Romanovsky","email":"vromanso@redhat.com","username":"vladikr"},"change_message_id":"4dbd450ef9ef32c386ae800bf6ea436d5ba083c4","unresolved":false,"context_lines":[{"line_number":92,"context_line":"instance."},{"line_number":93,"context_line":""},{"line_number":94,"context_line":"    nova boot -nic net_id\u003d\u003cid\u003e,queues\u003d#"},{"line_number":95,"context_line":"    nova interface-attach --net-id\u003d\u003cid\u003e --queues\u003d#"},{"line_number":96,"context_line":""},{"line_number":97,"context_line":"NOTE: Virtio-net multiqueue should be enabled in the guest OS manually, using"},{"line_number":98,"context_line":"ethtool."}],"source_content_type":"text/x-rst","patch_set":7,"id":"1a6ced46_afa01da0","line":95,"in_reply_to":"9a80dd14_526306dd","updated":"2015-03-24 17:33:19.000000000","message":"That was the initial argument. However, along the way it was clear that it will make it very difficult to handle, as for 10 different nics you would have to have 10 image-meta properties and correlate these to the vnics.","commit_id":"65b49f330669c11695f70ea744cd3e775fad9962"},{"author":{"_account_id":7730,"name":"Sahid Orentino Ferdjaoui","email":"sahid.ferdjaoui@industrialdiscipline.com","username":"sahid"},"change_message_id":"46bbbec4c5334df2501ed000ab4a9673178e8a3c","unresolved":false,"context_lines":[{"line_number":107,"context_line":""},{"line_number":108,"context_line":"None"},{"line_number":109,"context_line":""},{"line_number":110,"context_line":"REST API impact"},{"line_number":111,"context_line":"---------------"},{"line_number":112,"context_line":""},{"line_number":113,"context_line":"None."}],"source_content_type":"text/x-rst","patch_set":7,"id":"9a80dd14_979ad06c","line":110,"updated":"2015-03-17 10:33:19.000000000","message":"Since you need to update the boot and interface-attach API, don\u0027t you have to add a note here?","commit_id":"65b49f330669c11695f70ea744cd3e775fad9962"},{"author":{"_account_id":7730,"name":"Sahid Orentino Ferdjaoui","email":"sahid.ferdjaoui@industrialdiscipline.com","username":"sahid"},"change_message_id":"8ffd03eadd56a007f4d54f64db065797d0c163d3","unresolved":false,"context_lines":[{"line_number":107,"context_line":""},{"line_number":108,"context_line":"None"},{"line_number":109,"context_line":""},{"line_number":110,"context_line":"REST API impact"},{"line_number":111,"context_line":"---------------"},{"line_number":112,"context_line":""},{"line_number":113,"context_line":"None."}],"source_content_type":"text/x-rst","patch_set":7,"id":"1a6ced46_92caa66c","line":110,"in_reply_to":"1a6ced46_4f8889e7","updated":"2015-03-25 11:15:58.000000000","message":"Yes i was expected a formatted json as an example","commit_id":"65b49f330669c11695f70ea744cd3e775fad9962"},{"author":{"_account_id":8802,"name":"Vladik Romanovsky","email":"vromanso@redhat.com","username":"vladikr"},"change_message_id":"4dbd450ef9ef32c386ae800bf6ea436d5ba083c4","unresolved":false,"context_lines":[{"line_number":107,"context_line":""},{"line_number":108,"context_line":"None"},{"line_number":109,"context_line":""},{"line_number":110,"context_line":"REST API impact"},{"line_number":111,"context_line":"---------------"},{"line_number":112,"context_line":""},{"line_number":113,"context_line":"None."}],"source_content_type":"text/x-rst","patch_set":7,"id":"1a6ced46_4f8889e7","line":110,"in_reply_to":"9a80dd14_979ad06c","updated":"2015-03-24 17:33:19.000000000","message":"I\u0027m not planing to change the REST API structure. The -nic creates a json blob that is being handled by the NetworkRequest object.","commit_id":"65b49f330669c11695f70ea744cd3e775fad9962"}]}
