)]}'
{"/COMMIT_MSG":[{"author":{"_account_id":782,"name":"John Garbutt","email":"john@johngarbutt.com","username":"johngarbutt"},"change_message_id":"5aa6821270491c1cf1c708e21e684c7726292f5f","unresolved":false,"context_lines":[{"line_number":15,"context_line":""},{"line_number":16,"context_line":"* Provides L2 redundancy when multiple devices are allocated on behalf"},{"line_number":17,"context_line":"  of a single guest."},{"line_number":18,"context_line":""},{"line_number":19,"context_line":"Change-Id: I4af2a35de9ece709b774df9c05da33fef47157db"}],"source_content_type":"text/x-gerrit-commit-message","patch_set":3,"id":"9a0c5dc1_539c1a69","line":18,"updated":"2015-05-12 13:20:39.000000000","message":"Please add (and create) the blueprint, I guess:\n\nblueprint distribute-pci-allocation","commit_id":"4e7f14a84903978337fa4b42a382b68f3345f901"}],"specs/kilo/approved/distribute-pci-allocations.rst":[{"author":{"_account_id":7,"name":"Jay Pipes","email":"jaypipes@gmail.com","username":"jaypipes"},"change_message_id":"ba162d5422fb507b54e3fc5c93c9b10017ef4bc8","unresolved":false,"context_lines":[{"line_number":8,"context_line":"Distribute PCI Requests Across Multiple Devices"},{"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\u003d\u003d\u003d\u003d\u003d"},{"line_number":10,"context_line":""},{"line_number":11,"context_line":"Include the URL of your launchpad blueprint:"},{"line_number":12,"context_line":""},{"line_number":13,"context_line":"https://blueprints.launchpad.net/nova/+spec/distribute-pci-allocation.rst"},{"line_number":14,"context_line":""}],"source_content_type":"text/x-rst","patch_set":1,"id":"3a961159_d46589af","line":11,"updated":"2014-12-22 21:08:33.000000000","message":"You can remove the above line :)","commit_id":"58cae5e806fb1c1f1e5091022dff965621172c6b"},{"author":{"_account_id":6685,"name":"Baodong (Robert) Li","email":"baoli@cisco.com","username":"baoli"},"change_message_id":"bc71df2f3c27da73f825f1917d3aff77bfbf115e","unresolved":false,"context_lines":[{"line_number":35,"context_line":""},{"line_number":36,"context_line":"Creating an Openstack instance with multiple SR-IOV ports where the ports are"},{"line_number":37,"context_line":"instantiated across diverse physical links provides more even device and link"},{"line_number":38,"context_line":"utilisation and allows guests to take advantage of L2 redundant links for"},{"line_number":39,"context_line":"bonding, etc."},{"line_number":40,"context_line":""},{"line_number":41,"context_line":"Project Priority"}],"source_content_type":"text/x-rst","patch_set":1,"id":"3a961159_c9f8c44c","line":38,"updated":"2015-01-06 12:52:27.000000000","message":"A physical port may have multiple physical functions. In that case, distributing the requests among the physical functions will not provide the benefits as described in above.\n\nI think that we need to clarify the difference between \u0027links\u0027 and physical functions.","commit_id":"58cae5e806fb1c1f1e5091022dff965621172c6b"},{"author":{"_account_id":6681,"name":"Brent Eagles","email":"beagles@redhat.com","username":"beagles"},"change_message_id":"82fc33fc41a6b024c73150e95fc209b3d8e0bbcc","unresolved":false,"context_lines":[{"line_number":35,"context_line":""},{"line_number":36,"context_line":"Creating an Openstack instance with multiple SR-IOV ports where the ports are"},{"line_number":37,"context_line":"instantiated across diverse physical links provides more even device and link"},{"line_number":38,"context_line":"utilisation and allows guests to take advantage of L2 redundant links for"},{"line_number":39,"context_line":"bonding, etc."},{"line_number":40,"context_line":""},{"line_number":41,"context_line":"Project Priority"}],"source_content_type":"text/x-rst","patch_set":1,"id":"3a961159_3f55f6c1","line":38,"in_reply_to":"3a961159_c9f8c44c","updated":"2015-01-06 13:44:44.000000000","message":"NIC partitioning (NPAR) type of thing, right? Is there a way to discern physical links from the PCI information in a deterministic (and usable) way? With respect to clarification would it be sufficient to elaborate the conditions in which L2 redundancy would be obtained, i.e. where this is a single PF per physical ethernet port?","commit_id":"58cae5e806fb1c1f1e5091022dff965621172c6b"},{"author":{"_account_id":11647,"name":"Przemyslaw Czesnowicz","email":"przemyslaw.czesnowicz@intel.com","username":"pczesno"},"change_message_id":"f67c576f8df9f62d804c643bf2cf3e8c465a9ddd","unresolved":false,"context_lines":[{"line_number":55,"context_line":"* Extend the PCI device request API to support multiple requests, signifying"},{"line_number":56,"context_line":"  that the requests are on behalf of a single guest instance. This"},{"line_number":57,"context_line":"  implementation supports failure if the only available virtual functions are"},{"line_number":58,"context_line":"  on the same physical function."},{"line_number":59,"context_line":""},{"line_number":60,"context_line":"Alternatives"},{"line_number":61,"context_line":"------------"}],"source_content_type":"text/x-rst","patch_set":1,"id":"3a961159_fa8ffc20","line":58,"updated":"2015-01-06 14:50:32.000000000","message":"Would the scheduling always fail if available virtual functions come from the same physical function?\nIf so this might not be optimal. In many cases there is nothing wrong with using vf\u0027s from same pf.\nCan the spreading be done on a best effort basis, an only fail if the user signifies that he needs to use the vf\u0027s for bonding?","commit_id":"58cae5e806fb1c1f1e5091022dff965621172c6b"},{"author":{"_account_id":6681,"name":"Brent Eagles","email":"beagles@redhat.com","username":"beagles"},"change_message_id":"e97184cfa5f642daed8b4ba1fffa69250015e03a","unresolved":false,"context_lines":[{"line_number":55,"context_line":"* Extend the PCI device request API to support multiple requests, signifying"},{"line_number":56,"context_line":"  that the requests are on behalf of a single guest instance. This"},{"line_number":57,"context_line":"  implementation supports failure if the only available virtual functions are"},{"line_number":58,"context_line":"  on the same physical function."},{"line_number":59,"context_line":""},{"line_number":60,"context_line":"Alternatives"},{"line_number":61,"context_line":"------------"}],"source_content_type":"text/x-rst","patch_set":1,"id":"3a961159_23dec0da","line":58,"in_reply_to":"3a961159_fa8ffc20","updated":"2015-01-06 16:01:15.000000000","message":"For the cases where this is desired, silently going to the same PF might be considered a fail. Where this would be likely and \"advanced usage\", the best thing would be to make whether or not it is an error a property of the port.\n\nI\u0027m curious about the \"many cases\" you mention. Do you have examples?","commit_id":"58cae5e806fb1c1f1e5091022dff965621172c6b"},{"author":{"_account_id":11647,"name":"Przemyslaw Czesnowicz","email":"przemyslaw.czesnowicz@intel.com","username":"pczesno"},"change_message_id":"f67c576f8df9f62d804c643bf2cf3e8c465a9ddd","unresolved":false,"context_lines":[{"line_number":64,"context_line":"more even utilization. Ensuring multiple SR-IOV device requests for a guest"},{"line_number":65,"context_line":"span multiple physical links could employ additional port detail information"},{"line_number":66,"context_line":"but would still require PCI request scheduling changes."},{"line_number":67,"context_line":""},{"line_number":68,"context_line":"Data model impact"},{"line_number":69,"context_line":"-----------------"},{"line_number":70,"context_line":""}],"source_content_type":"text/x-rst","patch_set":1,"id":"3a961159_5a5c907d","line":67,"updated":"2015-01-06 14:50:32.000000000","message":"An alternative approach could be to schedule based on available throughput","commit_id":"58cae5e806fb1c1f1e5091022dff965621172c6b"},{"author":{"_account_id":6681,"name":"Brent Eagles","email":"beagles@redhat.com","username":"beagles"},"change_message_id":"e97184cfa5f642daed8b4ba1fffa69250015e03a","unresolved":false,"context_lines":[{"line_number":64,"context_line":"more even utilization. Ensuring multiple SR-IOV device requests for a guest"},{"line_number":65,"context_line":"span multiple physical links could employ additional port detail information"},{"line_number":66,"context_line":"but would still require PCI request scheduling changes."},{"line_number":67,"context_line":""},{"line_number":68,"context_line":"Data model impact"},{"line_number":69,"context_line":"-----------------"},{"line_number":70,"context_line":""}],"source_content_type":"text/x-rst","patch_set":1,"id":"3a961159_e3acb837","line":67,"in_reply_to":"3a961159_5a5c907d","updated":"2015-01-06 16:01:15.000000000","message":"True, if the only intent is to ensure performance. If the intent is to instigate distribution across multiple links for things like HA links, then throughput based scheduling is not an appropriate approach. That being said, it is worth mentioning if only to describe why it won\u0027t cut it.","commit_id":"58cae5e806fb1c1f1e5091022dff965621172c6b"}],"specs/liberty/approved/distribute-pci-allocations.rst":[{"author":{"_account_id":6509,"name":"Roman Bogorodskiy","email":"rbogorodskiy@mirantis.com","username":"novel"},"change_message_id":"03fb80f41997dab563a5330ba94532361c66236a","unresolved":false,"context_lines":[{"line_number":8,"context_line":"Distribute PCI Requests Across Multiple Devices"},{"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\u003d\u003d\u003d\u003d\u003d"},{"line_number":10,"context_line":""},{"line_number":11,"context_line":"https://blueprints.launchpad.net/nova/+spec/distribute-pci-allocation.rst"},{"line_number":12,"context_line":""},{"line_number":13,"context_line":"PCI requests are provisioned in a list based fashion. In SR-IOV networking"},{"line_number":14,"context_line":"devices, a set of candidate virtual functions can span multiple physical"}],"source_content_type":"text/x-rst","patch_set":3,"id":"9a0c5dc1_fccf6db1","line":11,"updated":"2015-05-12 06:55:50.000000000","message":"Is that link broken?","commit_id":"4e7f14a84903978337fa4b42a382b68f3345f901"},{"author":{"_account_id":6681,"name":"Brent Eagles","email":"beagles@redhat.com","username":"beagles"},"change_message_id":"aadd396941937d7b7c3c1e73828d72d7cd14a742","unresolved":false,"context_lines":[{"line_number":8,"context_line":"Distribute PCI Requests Across Multiple Devices"},{"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\u003d\u003d\u003d\u003d\u003d"},{"line_number":10,"context_line":""},{"line_number":11,"context_line":"https://blueprints.launchpad.net/nova/+spec/distribute-pci-allocation.rst"},{"line_number":12,"context_line":""},{"line_number":13,"context_line":"PCI requests are provisioned in a list based fashion. In SR-IOV networking"},{"line_number":14,"context_line":"devices, a set of candidate virtual functions can span multiple physical"}],"source_content_type":"text/x-rst","patch_set":3,"id":"7a016987_3fa957e0","line":11,"in_reply_to":"9a0c5dc1_d3afea92","updated":"2015-05-26 18:35:40.000000000","message":"Gah! I was sure I had done so previously. I probably forgot to hit submit or whatever.","commit_id":"4e7f14a84903978337fa4b42a382b68f3345f901"},{"author":{"_account_id":782,"name":"John Garbutt","email":"john@johngarbutt.com","username":"johngarbutt"},"change_message_id":"5aa6821270491c1cf1c708e21e684c7726292f5f","unresolved":false,"context_lines":[{"line_number":8,"context_line":"Distribute PCI Requests Across Multiple Devices"},{"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\u003d\u003d\u003d\u003d\u003d"},{"line_number":10,"context_line":""},{"line_number":11,"context_line":"https://blueprints.launchpad.net/nova/+spec/distribute-pci-allocation.rst"},{"line_number":12,"context_line":""},{"line_number":13,"context_line":"PCI requests are provisioned in a list based fashion. In SR-IOV networking"},{"line_number":14,"context_line":"devices, a set of candidate virtual functions can span multiple physical"}],"source_content_type":"text/x-rst","patch_set":3,"id":"9a0c5dc1_d3afea92","line":11,"in_reply_to":"9a0c5dc1_fccf6db1","updated":"2015-05-12 13:20:39.000000000","message":"+1 please actually create the blueprint, I guess you mean:\nhttps://blueprints.launchpad.net/nova/+spec/distribute-pci-allocation","commit_id":"4e7f14a84903978337fa4b42a382b68f3345f901"},{"author":{"_account_id":782,"name":"John Garbutt","email":"john@johngarbutt.com","username":"johngarbutt"},"change_message_id":"5aa6821270491c1cf1c708e21e684c7726292f5f","unresolved":false,"context_lines":[{"line_number":35,"context_line":"an Openstack instance with multiple SR-IOV ports, spreading the port"},{"line_number":36,"context_line":"allocation across candidate physical functions provides more even device and"},{"line_number":37,"context_line":"link utilization as well as allowing guests to take advantage of L2 redundant"},{"line_number":38,"context_line":"links for bonding, etc."},{"line_number":39,"context_line":""},{"line_number":40,"context_line":"Project Priority"},{"line_number":41,"context_line":"-----------------"}],"source_content_type":"text/x-rst","patch_set":3,"id":"9a0c5dc1_33c76614","line":38,"updated":"2015-05-12 13:20:39.000000000","message":"I think we still only allow one nic per network attached to each VM, or did we relax that restriction already?","commit_id":"4e7f14a84903978337fa4b42a382b68f3345f901"},{"author":{"_account_id":7,"name":"Jay Pipes","email":"jaypipes@gmail.com","username":"jaypipes"},"change_message_id":"53c4d7d246bd4fc9871e3a2ee3de7daeeb57d604","unresolved":false,"context_lines":[{"line_number":35,"context_line":"an Openstack instance with multiple SR-IOV ports, spreading the port"},{"line_number":36,"context_line":"allocation across candidate physical functions provides more even device and"},{"line_number":37,"context_line":"link utilization as well as allowing guests to take advantage of L2 redundant"},{"line_number":38,"context_line":"links for bonding, etc."},{"line_number":39,"context_line":""},{"line_number":40,"context_line":"Project Priority"},{"line_number":41,"context_line":"-----------------"}],"source_content_type":"text/x-rst","patch_set":3,"id":"7a016987_f545e6b2","line":38,"in_reply_to":"9a0c5dc1_33c76614","updated":"2015-05-17 00:23:13.000000000","message":"No, John, I don\u0027t see that restriction any more. nova.objects.NetworkRequest is essentially a tuple of (network_id, address, port_id, pci_request_id), with all fields in the tuple being nullable. This seems to indicate no such restriction. Furthermore, checking the code in the server controller\u0027s _get_requested_networks() function:\n\nhttps://github.com/openstack/nova/blob/master/nova/api/openstack/compute/servers.py#L290\n\nit doesn\u0027t seem that there are any such restrictions if Neutron v2 is used:\n\nhttps://github.com/openstack/nova/blob/master/nova/api/openstack/compute/servers.py#L331-L337","commit_id":"4e7f14a84903978337fa4b42a382b68f3345f901"},{"author":{"_account_id":7,"name":"Jay Pipes","email":"jaypipes@gmail.com","username":"jaypipes"},"change_message_id":"53c4d7d246bd4fc9871e3a2ee3de7daeeb57d604","unresolved":false,"context_lines":[{"line_number":40,"context_line":"Project Priority"},{"line_number":41,"context_line":"-----------------"},{"line_number":42,"context_line":""},{"line_number":43,"context_line":"Does not directly address listed Kilo priorities."},{"line_number":44,"context_line":""},{"line_number":45,"context_line":"Proposed change"},{"line_number":46,"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":3,"id":"7a016987_f51cc6c2","line":43,"updated":"2015-05-17 00:23:13.000000000","message":"s/Kilo/Liberty/","commit_id":"4e7f14a84903978337fa4b42a382b68f3345f901"},{"author":{"_account_id":782,"name":"John Garbutt","email":"john@johngarbutt.com","username":"johngarbutt"},"change_message_id":"5aa6821270491c1cf1c708e21e684c7726292f5f","unresolved":false,"context_lines":[{"line_number":47,"context_line":""},{"line_number":48,"context_line":"The proposed change is twofold:"},{"line_number":49,"context_line":""},{"line_number":50,"context_line":"* Alter the PCI device request scheduling to, in the SR-IOV case, distributed"},{"line_number":51,"context_line":"  PCI device requests evenly across available multiple physical functions for"},{"line_number":52,"context_line":"  the same associated label."},{"line_number":53,"context_line":""}],"source_content_type":"text/x-rst","patch_set":3,"id":"9a0c5dc1_2efffd02","line":50,"updated":"2015-05-12 13:20:39.000000000","message":"nit: s/distributed/distribute/","commit_id":"4e7f14a84903978337fa4b42a382b68f3345f901"},{"author":{"_account_id":7543,"name":"Yongli He","email":"yongli.he@intel.com","username":"yongli.he"},"change_message_id":"c3d29d63127ec1e87d31bc4147ae2a6c10a3744a","unresolved":false,"context_lines":[{"line_number":51,"context_line":"  PCI device requests evenly across available multiple physical functions for"},{"line_number":52,"context_line":"  the same associated label."},{"line_number":53,"context_line":""},{"line_number":54,"context_line":"* Extend the PCI device request API to support multiple requests, signifying"},{"line_number":55,"context_line":"  that the requests are on behalf of a single guest instance. This"},{"line_number":56,"context_line":"  implementation supports failure if the only available virtual functions are"},{"line_number":57,"context_line":"  on the same physical function."}],"source_content_type":"text/x-rst","patch_set":3,"id":"7a016987_cee18500","line":54,"updated":"2015-05-14 13:22:38.000000000","message":"i suppose this is the alias and pci request related, if so, there support request multiple device already. maybe here you want to say like \"got way to request cross physical device\" ? fix me.","commit_id":"4e7f14a84903978337fa4b42a382b68f3345f901"},{"author":{"_account_id":7,"name":"Jay Pipes","email":"jaypipes@gmail.com","username":"jaypipes"},"change_message_id":"53c4d7d246bd4fc9871e3a2ee3de7daeeb57d604","unresolved":false,"context_lines":[{"line_number":53,"context_line":""},{"line_number":54,"context_line":"* Extend the PCI device request API to support multiple requests, signifying"},{"line_number":55,"context_line":"  that the requests are on behalf of a single guest instance. This"},{"line_number":56,"context_line":"  implementation supports failure if the only available virtual functions are"},{"line_number":57,"context_line":"  on the same physical function."},{"line_number":58,"context_line":""},{"line_number":59,"context_line":"Alternatives"}],"source_content_type":"text/x-rst","patch_set":3,"id":"7a016987_1579a2f6","line":56,"updated":"2015-05-17 00:23:13.000000000","message":"Note that it\u0027s not actually multiple PCI requests. Rather, it\u0027s a single request for multiple VFs, distributed across multiple PFs. It\u0027s not just a semantic difference; the difference indicates that the two VFs should be placed in a single atomic request, or not at all.","commit_id":"4e7f14a84903978337fa4b42a382b68f3345f901"},{"author":{"_account_id":6681,"name":"Brent Eagles","email":"beagles@redhat.com","username":"beagles"},"change_message_id":"aadd396941937d7b7c3c1e73828d72d7cd14a742","unresolved":false,"context_lines":[{"line_number":53,"context_line":""},{"line_number":54,"context_line":"* Extend the PCI device request API to support multiple requests, signifying"},{"line_number":55,"context_line":"  that the requests are on behalf of a single guest instance. This"},{"line_number":56,"context_line":"  implementation supports failure if the only available virtual functions are"},{"line_number":57,"context_line":"  on the same physical function."},{"line_number":58,"context_line":""},{"line_number":59,"context_line":"Alternatives"}],"source_content_type":"text/x-rst","patch_set":3,"id":"7a016987_5f59b38d","line":56,"in_reply_to":"7a016987_1579a2f6","updated":"2015-05-26 18:35:40.000000000","message":"Right. I think I\u0027ve captured this in the next update.","commit_id":"4e7f14a84903978337fa4b42a382b68f3345f901"},{"author":{"_account_id":782,"name":"John Garbutt","email":"john@johngarbutt.com","username":"johngarbutt"},"change_message_id":"5aa6821270491c1cf1c708e21e684c7726292f5f","unresolved":false,"context_lines":[{"line_number":54,"context_line":"* Extend the PCI device request API to support multiple requests, signifying"},{"line_number":55,"context_line":"  that the requests are on behalf of a single guest instance. This"},{"line_number":56,"context_line":"  implementation supports failure if the only available virtual functions are"},{"line_number":57,"context_line":"  on the same physical function."},{"line_number":58,"context_line":""},{"line_number":59,"context_line":"Alternatives"},{"line_number":60,"context_line":"------------"}],"source_content_type":"text/x-rst","patch_set":3,"id":"9a0c5dc1_2e749d9a","line":57,"updated":"2015-05-12 13:20:39.000000000","message":"Nit: I am guessing this this is an internal API right? Maybe worth saying that explicitly. Maybe thats overkill.","commit_id":"4e7f14a84903978337fa4b42a382b68f3345f901"},{"author":{"_account_id":782,"name":"John Garbutt","email":"john@johngarbutt.com","username":"johngarbutt"},"change_message_id":"5aa6821270491c1cf1c708e21e684c7726292f5f","unresolved":false,"context_lines":[{"line_number":67,"context_line":"Link optimization could be improved by providing scheduling based on available"},{"line_number":68,"context_line":"throughput, However, this does not satisfy the physical link redundancy use"},{"line_number":69,"context_line":"case. Also, it would only work where throughput details are actually provided"},{"line_number":70,"context_line":"by the underlying device."},{"line_number":71,"context_line":""},{"line_number":72,"context_line":"Data model impact"},{"line_number":73,"context_line":"-----------------"}],"source_content_type":"text/x-rst","patch_set":3,"id":"9a0c5dc1_2e37fdb1","line":70,"updated":"2015-05-12 13:20:39.000000000","message":"QoS rules might come into play here I guess? but not for this version, thats fine. probably not worth adding.","commit_id":"4e7f14a84903978337fa4b42a382b68f3345f901"},{"author":{"_account_id":6681,"name":"Brent Eagles","email":"beagles@redhat.com","username":"beagles"},"change_message_id":"aadd396941937d7b7c3c1e73828d72d7cd14a742","unresolved":false,"context_lines":[{"line_number":67,"context_line":"Link optimization could be improved by providing scheduling based on available"},{"line_number":68,"context_line":"throughput, However, this does not satisfy the physical link redundancy use"},{"line_number":69,"context_line":"case. Also, it would only work where throughput details are actually provided"},{"line_number":70,"context_line":"by the underlying device."},{"line_number":71,"context_line":""},{"line_number":72,"context_line":"Data model impact"},{"line_number":73,"context_line":"-----------------"}],"source_content_type":"text/x-rst","patch_set":3,"id":"7a016987_5fd093c9","line":70,"in_reply_to":"9a0c5dc1_2e37fdb1","updated":"2015-05-26 18:35:40.000000000","message":"Yeah, I\u0027m shying away from control mechanisms for now.","commit_id":"4e7f14a84903978337fa4b42a382b68f3345f901"},{"author":{"_account_id":6509,"name":"Roman Bogorodskiy","email":"rbogorodskiy@mirantis.com","username":"novel"},"change_message_id":"787aa16ba62816aa05a984d7a58eeae26de3e79f","unresolved":false,"context_lines":[{"line_number":8,"context_line":"Distribute PCI Requests Across Multiple Devices"},{"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\u003d\u003d\u003d\u003d\u003d"},{"line_number":10,"context_line":""},{"line_number":11,"context_line":"https://blueprints.launchpad.net/nova/+spec/distribute-pci-allocation.rst"},{"line_number":12,"context_line":""},{"line_number":13,"context_line":"PCI requests are provisioned in a list based fashion. In SR-IOV networking"},{"line_number":14,"context_line":"devices, a set of candidate virtual functions can span multiple physical"}],"source_content_type":"text/x-rst","patch_set":4,"id":"3afb71cf_145d0479","line":11,"updated":"2015-06-09 07:13:12.000000000","message":"\u0027s/\\.rst$//\u0027 to get the right URL","commit_id":"6a0365849f0bc1c9bdd2952c0aef729c1ceaa1e0"},{"author":{"_account_id":782,"name":"John Garbutt","email":"john@johngarbutt.com","username":"johngarbutt"},"change_message_id":"006f8d92487865019c7f406b1a24703c4164f7ba","unresolved":false,"context_lines":[{"line_number":51,"context_line":"  consumption of VFs evenly across available multiple physical functions for"},{"line_number":52,"context_line":"  the same associated label."},{"line_number":53,"context_line":""},{"line_number":54,"context_line":"* Add an option to the the PCI device request API to force a failure if a"},{"line_number":55,"context_line":"  request for multiple VFs cannot be distributed across separate PFs then the"},{"line_number":56,"context_line":"  request should fail."},{"line_number":57,"context_line":""}],"source_content_type":"text/x-rst","patch_set":4,"id":"3afb71cf_25d4771e","line":54,"updated":"2015-06-08 12:29:32.000000000","message":"Sorry, how does that \"option\" work here? Is that a nova.conf value or something the user can tweak?\n\nI am wondering if we should default this value to True, and maybe drop the option all together (only add it if someone requests it). It seems like 99% of the requests that come in what to be spread, and probably want to fail if thats not possible.","commit_id":"6a0365849f0bc1c9bdd2952c0aef729c1ceaa1e0"},{"author":{"_account_id":6681,"name":"Brent Eagles","email":"beagles@redhat.com","username":"beagles"},"change_message_id":"60d6f0c85398c45b6b74cc8a1489f8fdcfb035d6","unresolved":false,"context_lines":[{"line_number":51,"context_line":"  consumption of VFs evenly across available multiple physical functions for"},{"line_number":52,"context_line":"  the same associated label."},{"line_number":53,"context_line":""},{"line_number":54,"context_line":"* Add an option to the the PCI device request API to force a failure if a"},{"line_number":55,"context_line":"  request for multiple VFs cannot be distributed across separate PFs then the"},{"line_number":56,"context_line":"  request should fail."},{"line_number":57,"context_line":""}],"source_content_type":"text/x-rst","patch_set":4,"id":"3afb71cf_e94963b8","line":54,"in_reply_to":"3afb71cf_25d4771e","updated":"2015-06-09 13:34:06.000000000","message":"I was thinking of something that would be set in the vif details for the port, but I tend to agree with you. That is better to assume distributing is needed. The case is so specific - both vfs on the same network - that it wouldn\u0027t effect most other users. I\u0027ll alter the lingo to omit the option, but specify failing if the distribution isn\u0027t possible.","commit_id":"6a0365849f0bc1c9bdd2952c0aef729c1ceaa1e0"},{"author":{"_account_id":6509,"name":"Roman Bogorodskiy","email":"rbogorodskiy@mirantis.com","username":"novel"},"change_message_id":"7cb56f933c775f9ae55de01be74126d3c61f498e","unresolved":false,"context_lines":[{"line_number":51,"context_line":"  consumption of VFs evenly across available multiple physical functions for"},{"line_number":52,"context_line":"  the same associated label."},{"line_number":53,"context_line":""},{"line_number":54,"context_line":"* Add an option to the the PCI device request API to force a failure if a"},{"line_number":55,"context_line":"  request for multiple VFs cannot be distributed across separate PFs then the"},{"line_number":56,"context_line":"  request should fail."},{"line_number":57,"context_line":""}],"source_content_type":"text/x-rst","patch_set":4,"id":"3afb71cf_147a99bd","line":54,"in_reply_to":"3afb71cf_25d4771e","updated":"2015-06-09 11:47:22.000000000","message":"If we\u0027re talking about nova.conf option, I think it could be a little confusing for users. Also, it doesn\u0027t seem to be flexible because that\u0027ll affect an entire deployment.\n\nProbably it\u0027d be more straight-forward to fail if VFs cannot be distributed, just like you set, and allow to override that using image properties if user really needs that.","commit_id":"6a0365849f0bc1c9bdd2952c0aef729c1ceaa1e0"},{"author":{"_account_id":782,"name":"John Garbutt","email":"john@johngarbutt.com","username":"johngarbutt"},"change_message_id":"ea790aac0573e218f1da7d0358c2cbc504a09f48","unresolved":false,"context_lines":[{"line_number":77,"context_line":"REST API impact"},{"line_number":78,"context_line":"---------------"},{"line_number":79,"context_line":""},{"line_number":80,"context_line":"The implementation is an internal change only and has no impact."},{"line_number":81,"context_line":""},{"line_number":82,"context_line":"Security impact"},{"line_number":83,"context_line":"---------------"}],"source_content_type":"text/x-rst","patch_set":4,"id":"3afb71cf_25891727","line":80,"updated":"2015-06-08 12:30:22.000000000","message":"Not totally sure thats the correct choice. Maybe we should fail the build if we can\u0027t spread between multiple physical networks?","commit_id":"6a0365849f0bc1c9bdd2952c0aef729c1ceaa1e0"},{"author":{"_account_id":6685,"name":"Baodong (Robert) Li","email":"baoli@cisco.com","username":"baoli"},"change_message_id":"409d57aca6d782021f8f4f339ff4a80cd4675ddb","unresolved":false,"context_lines":[{"line_number":31,"context_line":"Use Cases"},{"line_number":32,"context_line":"----------"},{"line_number":33,"context_line":""},{"line_number":34,"context_line":"On systems where there is a physical function per physical link and creating"},{"line_number":35,"context_line":"an Openstack instance with multiple SR-IOV ports, spreading the port"},{"line_number":36,"context_line":"allocation across candidate physical functions provides more even device and"},{"line_number":37,"context_line":"link utilization as well as allowing guests to take advantage of L2 redundant"}],"source_content_type":"text/x-rst","patch_set":5,"id":"3afb71cf_24f052c9","line":34,"updated":"2015-06-11 13:43:30.000000000","message":"It looks like that redundancy can be achieved when a physical link has only one PF configured. Do we need a config option to turn this feature on/off in the case where a physical link contains multiple PFs?\n\nI have been trying to find out if it\u0027s possible to discover the associated physical link for a SR-IOV device on linux system. It would be useful for this feature.","commit_id":"e83ae7668c166137f80de2b72c8cf6d18a99ac9c"},{"author":{"_account_id":6509,"name":"Roman Bogorodskiy","email":"rbogorodskiy@mirantis.com","username":"novel"},"change_message_id":"675bb58c11f5e918aa264a755e829ad0b395882b","unresolved":false,"context_lines":[{"line_number":31,"context_line":"Use Cases"},{"line_number":32,"context_line":"----------"},{"line_number":33,"context_line":""},{"line_number":34,"context_line":"On systems where there is a physical function per physical link and creating"},{"line_number":35,"context_line":"an Openstack instance with multiple SR-IOV ports, spreading the port"},{"line_number":36,"context_line":"allocation across candidate physical functions provides more even device and"},{"line_number":37,"context_line":"link utilization as well as allowing guests to take advantage of L2 redundant"}],"source_content_type":"text/x-rst","patch_set":5,"id":"3afb71cf_4a08f72d","line":34,"in_reply_to":"3afb71cf_24f052c9","updated":"2015-06-11 14:32:58.000000000","message":"Yeah, it is possible. For VF with PCI addr $pci_addr \u0027/sys/bus/pci/devices/$pci_addr/physfn\u0027 would be a link to the associated physical link.","commit_id":"e83ae7668c166137f80de2b72c8cf6d18a99ac9c"},{"author":{"_account_id":6685,"name":"Baodong (Robert) Li","email":"baoli@cisco.com","username":"baoli"},"change_message_id":"1275bc42c9aff9f58f8b1d33fcccac4506b0641a","unresolved":false,"context_lines":[{"line_number":31,"context_line":"Use Cases"},{"line_number":32,"context_line":"----------"},{"line_number":33,"context_line":""},{"line_number":34,"context_line":"On systems where there is a physical function per physical link and creating"},{"line_number":35,"context_line":"an Openstack instance with multiple SR-IOV ports, spreading the port"},{"line_number":36,"context_line":"allocation across candidate physical functions provides more even device and"},{"line_number":37,"context_line":"link utilization as well as allowing guests to take advantage of L2 redundant"}],"source_content_type":"text/x-rst","patch_set":5,"id":"3afb71cf_2b5eda12","line":34,"in_reply_to":"3afb71cf_4a08f72d","updated":"2015-06-11 16:19:42.000000000","message":"that\u0027s pointing to the physical function (not the physical link).","commit_id":"e83ae7668c166137f80de2b72c8cf6d18a99ac9c"},{"author":{"_account_id":13997,"name":"SHIGEMATSU Mitsuhiro","email":"shigematsu.mitsuhiro@lab.ntt.co.jp","username":"pshige"},"change_message_id":"d68c5e3a494428578149eda971980d3781b1cfd8","unresolved":false,"context_lines":[{"line_number":48,"context_line":"The proposed change is twofold:"},{"line_number":49,"context_line":""},{"line_number":50,"context_line":"* The proposed change alters the PCI device request scheduling for PCI SR-IOV"},{"line_number":51,"context_line":"  to to distribute consumption of VFs evenly across available multiple physical"},{"line_number":52,"context_line":"  functions for the same associated label."},{"line_number":53,"context_line":""},{"line_number":54,"context_line":"* If a request involving multiple VFs with the same label occurs and a compute"}],"source_content_type":"text/x-rst","patch_set":5,"id":"da20952f_1aeb74dd","line":51,"updated":"2015-09-06 05:15:50.000000000","message":"s/to to/to/","commit_id":"e83ae7668c166137f80de2b72c8cf6d18a99ac9c"},{"author":{"_account_id":5511,"name":"Nikola Dipanov","email":"ndipanov@redhat.com","username":"ndipanov"},"change_message_id":"20e4c61135216cb57455167be903faf97055617f","unresolved":false,"context_lines":[{"line_number":54,"context_line":"* If a request involving multiple VFs with the same label occurs and a compute"},{"line_number":55,"context_line":"  node cannot distribute the request across multiple physical functions, the"},{"line_number":56,"context_line":"  compute node is not selected. If none of the available compute nodes are"},{"line_number":57,"context_line":"  selected, scheduling will fail with a NoValidHost exception."},{"line_number":58,"context_line":""},{"line_number":59,"context_line":"Alternatives"},{"line_number":60,"context_line":"------------"}],"source_content_type":"text/x-rst","patch_set":5,"id":"1af86dd1_aaee2bea","line":57,"updated":"2015-06-12 15:58:01.000000000","message":"Hmm so am I reading this right that a host with a single PF will now always be rejected for an instance that requires several VFs? That sounds too extreme surely we may want to allow for this.\n\nAlso how does this fit in with the NUMA node constraints we may have - which takes precedence (NUMA locality or load spread) and why.","commit_id":"e83ae7668c166137f80de2b72c8cf6d18a99ac9c"},{"author":{"_account_id":6509,"name":"Roman Bogorodskiy","email":"rbogorodskiy@mirantis.com","username":"novel"},"change_message_id":"e6013f6206866b586854de12dffa357c641ae833","unresolved":false,"context_lines":[{"line_number":54,"context_line":"* If a request involving multiple VFs with the same label occurs and a compute"},{"line_number":55,"context_line":"  node cannot distribute the request across multiple physical functions, the"},{"line_number":56,"context_line":"  compute node is not selected. If none of the available compute nodes are"},{"line_number":57,"context_line":"  selected, scheduling will fail with a NoValidHost exception."},{"line_number":58,"context_line":""},{"line_number":59,"context_line":"Alternatives"},{"line_number":60,"context_line":"------------"}],"source_content_type":"text/x-rst","patch_set":5,"id":"1af86dd1_cdf2a19a","line":57,"in_reply_to":"1af86dd1_aaee2bea","updated":"2015-06-12 16:19:29.000000000","message":"Nikola, do you have some practical examples in mind where user would want more than one VF on a single PF for a single instance? If there would be two PFs, it makes sense for failover, but for a single PF / single link it doesn\u0027t sound like a very common situation.","commit_id":"e83ae7668c166137f80de2b72c8cf6d18a99ac9c"},{"author":{"_account_id":11647,"name":"Przemyslaw Czesnowicz","email":"przemyslaw.czesnowicz@intel.com","username":"pczesno"},"change_message_id":"0aa8f180598d69359d4d0b04650ba9d0393eef75","unresolved":false,"context_lines":[{"line_number":54,"context_line":"* If a request involving multiple VFs with the same label occurs and a compute"},{"line_number":55,"context_line":"  node cannot distribute the request across multiple physical functions, the"},{"line_number":56,"context_line":"  compute node is not selected. If none of the available compute nodes are"},{"line_number":57,"context_line":"  selected, scheduling will fail with a NoValidHost exception."},{"line_number":58,"context_line":""},{"line_number":59,"context_line":"Alternatives"},{"line_number":60,"context_line":"------------"}],"source_content_type":"text/x-rst","patch_set":5,"id":"5a2ca52d_a33a18cc","line":57,"in_reply_to":"1af86dd1_cdf2a19a","updated":"2015-10-13 14:17:39.000000000","message":"One example could be when the two PF\u0027s in the system are affinitized to different numa nodes, if a user requests two VF\u0027s for a VM that has a single numa node topology, then this would imply that there would be a performance impact as the packets from one of the VF\u0027s would need to traverse the QPI to be processed in the VM.","commit_id":"e83ae7668c166137f80de2b72c8cf6d18a99ac9c"},{"author":{"_account_id":5511,"name":"Nikola Dipanov","email":"ndipanov@redhat.com","username":"ndipanov"},"change_message_id":"778cb049461b8ba6443e8d10a7801a7243c3544e","unresolved":false,"context_lines":[{"line_number":54,"context_line":"* If a request involving multiple VFs with the same label occurs and a compute"},{"line_number":55,"context_line":"  node cannot distribute the request across multiple physical functions, the"},{"line_number":56,"context_line":"  compute node is not selected. If none of the available compute nodes are"},{"line_number":57,"context_line":"  selected, scheduling will fail with a NoValidHost exception."},{"line_number":58,"context_line":""},{"line_number":59,"context_line":"Alternatives"},{"line_number":60,"context_line":"------------"}],"source_content_type":"text/x-rst","patch_set":5,"id":"5a2ca52d_0879aafc","line":57,"in_reply_to":"1af86dd1_cdf2a19a","updated":"2015-10-13 13:52:01.000000000","message":"Sorry for responding with a delay of a few months (lol). I think it will make testing needlessly difficult","commit_id":"e83ae7668c166137f80de2b72c8cf6d18a99ac9c"}],"specs/mitaka/approved/distribute-pci-allocations.rst":[{"author":{"_account_id":782,"name":"John Garbutt","email":"john@johngarbutt.com","username":"johngarbutt"},"change_message_id":"4485a66109e382b47931ef032cf7d16cbac75f4d","unresolved":false,"context_lines":[{"line_number":40,"context_line":"Project Priority"},{"line_number":41,"context_line":"-----------------"},{"line_number":42,"context_line":""},{"line_number":43,"context_line":"Does not directly address listed Mitaka priorities."},{"line_number":44,"context_line":""},{"line_number":45,"context_line":"Proposed change"},{"line_number":46,"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":6,"id":"da85f559_dc78e2c6","line":43,"updated":"2015-11-10 16:00:35.000000000","message":"Nit: we should drop this whole section now.","commit_id":"85f97a81874c1554f545dd82db6fed9f2e6005c3"},{"author":{"_account_id":6509,"name":"Roman Bogorodskiy","email":"rbogorodskiy@mirantis.com","username":"novel"},"change_message_id":"326d0f6d899d4650a98a757fc2d539be7b979c0e","unresolved":false,"context_lines":[{"line_number":40,"context_line":"Project Priority"},{"line_number":41,"context_line":"-----------------"},{"line_number":42,"context_line":""},{"line_number":43,"context_line":"Does not directly address listed Mitaka priorities."},{"line_number":44,"context_line":""},{"line_number":45,"context_line":"Proposed change"},{"line_number":46,"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":6,"id":"da85f559_032ede42","line":43,"in_reply_to":"da85f559_dc78e2c6","updated":"2015-11-11 11:37:59.000000000","message":"Done","commit_id":"85f97a81874c1554f545dd82db6fed9f2e6005c3"},{"author":{"_account_id":6697,"name":"Sandhya Dasu","email":"sandhya.openinfra@gmail.com","username":"sadasu"},"change_message_id":"ff7b427cb26e494b3a1a06b268fdc9800ec7ee71","unresolved":false,"context_lines":[{"line_number":62,"context_line":"but would still require PCI request scheduling changes."},{"line_number":63,"context_line":""},{"line_number":64,"context_line":"Link optimization could be improved by providing scheduling based on available"},{"line_number":65,"context_line":"throughput, However, this does not satisfy the physical link redundancy use"},{"line_number":66,"context_line":"case. Also, it would only work where throughput details are actually provided"},{"line_number":67,"context_line":"by the underlying device."},{"line_number":68,"context_line":""}],"source_content_type":"text/x-rst","patch_set":6,"id":"da85f559_30a34602","line":65,"updated":"2015-11-10 19:40:09.000000000","message":"Is there a problem supporting the physical link redundancy case or is it something no one is looking for right now?","commit_id":"85f97a81874c1554f545dd82db6fed9f2e6005c3"},{"author":{"_account_id":6509,"name":"Roman Bogorodskiy","email":"rbogorodskiy@mirantis.com","username":"novel"},"change_message_id":"f745b03b499a7519d1f48292aaf36338680c6128","unresolved":false,"context_lines":[{"line_number":62,"context_line":"but would still require PCI request scheduling changes."},{"line_number":63,"context_line":""},{"line_number":64,"context_line":"Link optimization could be improved by providing scheduling based on available"},{"line_number":65,"context_line":"throughput, However, this does not satisfy the physical link redundancy use"},{"line_number":66,"context_line":"case. Also, it would only work where throughput details are actually provided"},{"line_number":67,"context_line":"by the underlying device."},{"line_number":68,"context_line":""}],"source_content_type":"text/x-rst","patch_set":6,"id":"da85f559_4375c6f7","line":65,"in_reply_to":"da85f559_30a34602","updated":"2015-11-11 11:40:43.000000000","message":"Could you please elaborate on a case you\u0027re interested in?","commit_id":"85f97a81874c1554f545dd82db6fed9f2e6005c3"},{"author":{"_account_id":6697,"name":"Sandhya Dasu","email":"sandhya.openinfra@gmail.com","username":"sadasu"},"change_message_id":"98a22781e2b0420ee1667f207a354b1bb065573e","unresolved":false,"context_lines":[{"line_number":62,"context_line":"but would still require PCI request scheduling changes."},{"line_number":63,"context_line":""},{"line_number":64,"context_line":"Link optimization could be improved by providing scheduling based on available"},{"line_number":65,"context_line":"throughput, However, this does not satisfy the physical link redundancy use"},{"line_number":66,"context_line":"case. Also, it would only work where throughput details are actually provided"},{"line_number":67,"context_line":"by the underlying device."},{"line_number":68,"context_line":""}],"source_content_type":"text/x-rst","patch_set":6,"id":"da85f559_11f00dc8","line":65,"in_reply_to":"da85f559_4375c6f7","updated":"2015-11-17 19:20:45.000000000","message":"Not looking for a specific case. Since you mentioned it here, it would be nice if you specified with a little bit more details what this use case is, that you are not supporting.","commit_id":"85f97a81874c1554f545dd82db6fed9f2e6005c3"},{"author":{"_account_id":782,"name":"John Garbutt","email":"john@johngarbutt.com","username":"johngarbutt"},"change_message_id":"4485a66109e382b47931ef032cf7d16cbac75f4d","unresolved":false,"context_lines":[{"line_number":136,"context_line":"\u003d\u003d\u003d\u003d\u003d\u003d\u003d"},{"line_number":137,"context_line":""},{"line_number":138,"context_line":"The changes to the selection logic are testable through unit testing, as is"},{"line_number":139,"context_line":"the extension to the PCI request API."},{"line_number":140,"context_line":""},{"line_number":141,"context_line":"Documentation Impact"},{"line_number":142,"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":6,"id":"da85f559_7f2a0880","line":139,"updated":"2015-11-10 16:00:35.000000000","message":"I really want to see functional tests for this eventually.","commit_id":"85f97a81874c1554f545dd82db6fed9f2e6005c3"},{"author":{"_account_id":6509,"name":"Roman Bogorodskiy","email":"rbogorodskiy@mirantis.com","username":"novel"},"change_message_id":"326d0f6d899d4650a98a757fc2d539be7b979c0e","unresolved":false,"context_lines":[{"line_number":136,"context_line":"\u003d\u003d\u003d\u003d\u003d\u003d\u003d"},{"line_number":137,"context_line":""},{"line_number":138,"context_line":"The changes to the selection logic are testable through unit testing, as is"},{"line_number":139,"context_line":"the extension to the PCI request API."},{"line_number":140,"context_line":""},{"line_number":141,"context_line":"Documentation Impact"},{"line_number":142,"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":6,"id":"da85f559_e3b2bae4","line":139,"in_reply_to":"da85f559_7f2a0880","updated":"2015-11-11 11:37:59.000000000","message":"Done","commit_id":"85f97a81874c1554f545dd82db6fed9f2e6005c3"},{"author":{"_account_id":19212,"name":"Antonio López","email":"antonio.lopezgracia@telefonica.com","username":"alogra"},"change_message_id":"614b2d1e3b63b0ed577a8283fceb7535e41e383b","unresolved":false,"context_lines":[{"line_number":43,"context_line":"The proposed change alters the PCI device request scheduling for PCI SR-IOV"},{"line_number":44,"context_line":"to to distribute consumption of VFs evenly across available multiple physical"},{"line_number":45,"context_line":"functions for the same associated label."},{"line_number":46,"context_line":""},{"line_number":47,"context_line":"Distribution happens in best-effort fashion, so even if PCI request cannot"},{"line_number":48,"context_line":"be supported in a distributed fashion, but still could be satisfied by"},{"line_number":49,"context_line":"the current queue based allocation schema, a guest will be booted."}],"source_content_type":"text/x-rst","patch_set":7,"id":"ba8a016a_a43ba537","line":46,"updated":"2015-11-17 17:10:22.000000000","message":"Ideally the scheduling should take into account the available bandwidth in the PF (after other VFs have been cheduled) and the bandwith requested by the VF","commit_id":"49e21dbad96fad76a5adf98419b9a189e912507c"},{"author":{"_account_id":17822,"name":"Gideon Agmon","email":"gideon.agmon@nokia.com","username":"gideon59"},"change_message_id":"6f07c5750ecf87afe17ec33e1fcdc6202a0d874e","unresolved":false,"context_lines":[{"line_number":47,"context_line":"Distribution happens in best-effort fashion, so even if PCI request cannot"},{"line_number":48,"context_line":"be supported in a distributed fashion, but still could be satisfied by"},{"line_number":49,"context_line":"the current queue based allocation schema, a guest will be booted."},{"line_number":50,"context_line":""},{"line_number":51,"context_line":"Alternatives"},{"line_number":52,"context_line":"------------"},{"line_number":53,"context_line":""}],"source_content_type":"text/x-rst","patch_set":7,"id":"bab6814e_c449be90","line":50,"updated":"2016-05-18 05:07:59.000000000","message":"“best effort” allocation may not sufficient when the VM definitely requires redundancy. \nAlso, there may be 4 ports, 2 for dataplane and two for management/OAM using macvtap, all L3 protected, over two physical interfaces. So Nova must explicitly know about these pairs in order to properly distribute them. \nI think the above two arguments imply that the user has to explicitly define the anti-affinity between the specific ports, which must be imposed by Nova, as proposed by https://review.openstack.org/#/c/182242.","commit_id":"49e21dbad96fad76a5adf98419b9a189e912507c"},{"author":{"_account_id":6804,"name":"bruce-benjamin","email":"bruce.benjamin@jhuapl.edu","username":"bruce-benjamin"},"change_message_id":"2cd20755d164e89f095a39b8f11148ac6950363e","unresolved":false,"context_lines":[{"line_number":144,"context_line":"References"},{"line_number":145,"context_line":"\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d"},{"line_number":146,"context_line":""},{"line_number":147,"context_line":"None."}],"source_content_type":"text/x-rst","patch_set":7,"id":"ba8a016a_834ac924","line":147,"updated":"2015-11-18 22:42:45.000000000","message":"I\u0027d recommend at least one reference here.  A couple of suggestions are:    \n\nhttp://www.intel.com/content/www/us/en/pci-express/pci-sig-sr-iov-primer-sr-iov-technology-paper.html\n\nhttp://support.citrix.com/article/CTX126624","commit_id":"49e21dbad96fad76a5adf98419b9a189e912507c"}]}
