13:00:26 <baoli> #startmeeting PCI Passthrough
13:00:27 <openstack> Meeting started Tue Apr 29 13:00:26 2014 UTC and is due to finish in 60 minutes.  The chair is baoli. Information about MeetBot at http://wiki.debian.org/MeetBot.
13:00:28 <openstack> Useful Commands: #action #agreed #help #info #idea #link #topic #startvote.
13:00:30 <openstack> The meeting name has been set to 'pci_passthrough'
13:00:47 <baoli> Hi
13:01:18 <beagles> hi
13:01:25 <irenab> hi
13:01:27 <heyongli> hi
13:02:47 <baoli> Today, let's continue the bp's discussion. We need to save a few minutes to discuss about the summit
13:03:08 <heyongli> maybe summit first
13:03:10 <baoli> Irenab loaded a new patch for the BP: https://review.openstack.org/#/c/86606/5/specs/juno/pci-passthrough-sriov.rst
13:03:23 <baoli> heyongli, sure
13:04:35 <irenab> we have one slot of 40 mins to all PCI issues, no separate Yongli session
13:05:13 <heyongli> i saw that
13:05:34 <irenab> I think we should plan how to split the time to cover all issues we have
13:06:57 <irenab> heyongli: is it OK to focus on Sr-IOV networking first and then have general PCI issues you have to discuss?
13:07:12 <heyongli> fine to me
13:08:59 <irenab> baoli: any idea what to present on Sr-IOV networking part?
13:09:10 <irenab> shall it be over the nova-spec?
13:09:17 <heyongli> i quick overview the proposal change, but this broken the agreement about multi tag, i think it's needed. sriov need only physical_network, this does mean pci pass-through support should just use this.
13:09:54 <baoli> Irenab, yes
13:10:28 <irenab> We also have started etherpad: https://etherpad.openstack.org/p/pci_passthrough_cross_project
13:10:36 <baoli> heyongli, please check the alternative approach section.
13:11:09 <irenab> Do you want to meet on Monday to discuss stuff before the session?
13:11:22 <baoli> irenab, yes.
13:11:47 <baoli> Check comments from John: http://summit.openstack.org/cfp/details/248
13:12:01 <heyongli> i saw that,  i agree with sriov entry use this tag only, but can not force this is the only one tag the pci passthrough will support, if this is what irenab's intend, i agree.
13:12:10 <baoli> Let's ask him how he wants the session to be run as well
13:12:39 <irenab> baoli: great. Probably next meeting we can define the time slot, we may want to ask Jogh to join on Monday.
13:13:15 <irenab> baoli: +1 on consulting John
13:14:01 <irenab> before we dive into spec discussion, is it possible to have next meeting on Wed same time?
13:14:05 <baoli> Irenab, yes, let's ask him to join next week's IRC if he can. otherwise, we can start discussion with him througth email
13:14:22 <baoli> irenab, for next week's?
13:14:49 <irenab> only next week
13:15:00 <baoli> I'm fine with it.
13:15:41 <irenab> heyongli, beagles: is it ok for you?
13:15:57 <beagles> sounds good
13:16:21 <heyongli> next weeks Wed?
13:16:51 <irenab> Apr 30 instead of Apr 29
13:17:24 <irenab> Our holiday happens on Apr 29, I will  be out of work
13:17:41 <beagles> ?
13:17:59 <beagles> today is 29 :)
13:18:02 <irenab> My mistake, May 7 instead of May 6
13:18:03 <heyongli> tomorrow is vacation for me
13:18:17 <heyongli> May 7, fine to me
13:18:35 <baoli> ok, we'll have IRC on Wed next week
13:18:53 <irenab> thansk
13:19:22 <irenab> shall we discuss spec issues?
13:19:37 <baoli> #topic specs
13:20:08 <sadasu> for meeting change we just have to make sure we have a channel available at this time
13:20:26 <sadasu> irenab, will you be able to check that or do you want one of us to do it
13:21:04 <irenab> sadasu: I'll check and update on meeting wiki page
13:21:15 <sadasu> irenab: thanks :-)
13:21:35 <irenab> 2 issues I have regarding spec: high availability and live migration
13:22:04 <baoli> Irenab, the HA case needs to be clarified, and also what kind of change is needed
13:22:17 <irenab> live migration is not in, discussing with baoli decided to leave it to other bp.
13:23:08 <irenab> baoli: I am still not sure how to implement it, but seems that for embedded switch case there is not majic and VM should get 2 VFs for same virtual network
13:24:12 <irenab> I will try to update more information once have
13:24:32 <baoli> irenab, so VM needs to switch between the two VFs? On the other hand, we can't pass two nics from the same virtual network to a VM
13:24:52 <irenab> I know that currently it is not possible to boot VM with 2 vnics on same virtual network
13:25:16 <baoli> irenab, so yes, let's figure out more details offline
13:25:39 <irenab> baoli: maybe this is what should be changed, still need to investigate. The idea that there is possible to bond interfaces inside guest on 2 SR-IOV vNICs
13:26:41 <irenab> baoli: to overcome the limitation, may boot VM with one vNIC and then attach-interface on same virtual network
13:26:59 <heyongli> irenab, i post some comments to the nova spec.
13:27:10 <baoli> Irenab, how about IP addresses on the two vnics? are they going to be the same?
13:27:23 <irenab> heyongli: thanks.
13:28:15 <irenab> baoli: IP should be defined on bond interface, but I still do not have the complete picture of how to orchestrate it.
13:28:17 <heyongli> basically i'm fine with whole idea, but there are 2 things i think need to change
13:29:05 <heyongli> irenab, maybe the bond interface is what you talk about?
13:29:31 <irenab> heyongli: yes, but it should be bond in the guest
13:30:15 <baoli> irenab, let's leave the HA case offline for now.
13:30:21 <irenab> I would suggest to skip to other issues and come back to HA case later if we have time, as I understood from baoli, cisco has other alterntive that does not require 2 VFs
13:30:21 <heyongli> irenab, sure, anyway, there is no such thing 2 l3 interface have same ip, except it's bond l2 interface, is this right?
13:30:44 <irenab> heyongli: yes, bond l2 interfaces
13:30:51 <heyongli> we should discuss HA later,
13:31:01 <baoli> On live migration, we need volunteer to start the BP and the spec
13:31:46 <heyongli> and what kind of HA we talk about, the 2 nic to same network? or 2 nic to a HA external switch configuration? or both?
13:32:19 <irenab> heyongli: 2 nics same network
13:32:49 <heyongli> if no external ha switch configuration, how this will improve the HA?
13:33:16 <heyongli> 2 nics to same swich might not change too much for HA, right?
13:33:16 <irenab> regarding live migration there is very detailed description baoli already has, it can be used for live migration spec
13:33:37 <beagles> heyongli: was just thinking the same thing....
13:34:04 <heyongli> beagles, cool, this really matter to HA, i thinkk
13:34:21 <irenab> must mentioned that most of the new content in the spec is input from baoli, I am just sort of technical writer :-) to push it for review
13:34:51 <baoli> Irenab, thanks for pushing up the update.
13:35:15 <irenab> heyongli: I mean same virtual network
13:36:11 <irenab> baoli: you wellcome. Learned few things for rst format
13:36:38 <baoli> Ok, let's leave the HA case offline. Maybe it deserves another BP as well.
13:36:40 <heyongli> irenab, i know same virtual network, but physically HA is what i concern, same virtual network sit on same extremal physical network does not help too much.
13:37:09 <heyongli> baoli, sure, extra bp is cool for me.
13:37:21 <irenab> heyongli: I am not sure, but please let me provide more details later
13:37:37 <heyongli> irenab, sure
13:37:53 <heyongli> back to spec
13:38:03 <baoli> On this use case: Cloud Admin wants to disable PCI SR-IOV support for a compute node.
13:38:22 <baoli> Is this really needed?
13:40:15 <heyongli> baoli, maybe, but seems can be done with current nova,  get remove all vm, change configuration and restart this compute node service.
13:40:52 <heyongli> but this case not very clear,
13:41:12 <heyongli> and,  an sriov networking's stats entry should be <vendor_id, product_id, physcial_netowrk'.
13:41:33 <heyongli> request for sriov still can use the only key 'physical_network' as describe in the section PCI requests changes
13:41:41 <irenab> Let's say this node does not have VMs, and admin just wants to stop using it for hosting VMs with SR-IOV nics
13:41:55 <baoli> heyongli, if it's done through reconfiguring and restarting, it doens't need to be mentioned specifically in the spec.
13:42:04 <irenab> shall he remove white list and restart nova, or there is something else to do?
13:42:13 <heyongli> baoli, i also think so
13:42:46 <baoli> So, let's remove that line?
13:43:04 <irenab> heyongli, you added this case upon John request. To add case besides adding new node to existing cluster
13:43:19 <heyongli> that's request by John, maybe a clarify this is addressed now by this way
13:44:04 <irenab> let's try to get his unputs before removing it
13:44:11 <beagles> irenab, +1
13:44:54 <baoli> Okay, we'll get input from John on that one.
13:45:11 <baoli> Heyongli, back to your question about the stats key
13:45:30 <irenab> baoli: there are few comments heyongli posted on the spec. Will you have time to update the spec based on his comments?
13:46:14 <baoli> irenab, sure.
13:46:19 <heyongli> for add extra configuration``sriov_networking = phynet:auto``, i more prefer a extra moduli zed white-list, this almost same thing, maybe.
13:47:28 <heyongli> i mean, configuration pci device provisioning as : white-list, sriov-auto... even more ...
13:49:25 <heyongli> suggest add a modules .i,e 'ariov-auto' as another way to provisioning the pci devices, and these modules action like whitelist, filter some devices to pci module, and possible add extra tag to the device.
13:50:48 <heyongli> all these suggestion is update to the nova spec.
13:51:23 <irenab> heyongli: as alternatives? Can you please put this in a comment on the Spec?
13:52:06 <heyongli> already done as comments.
13:52:19 <irenab> heyongli: thanks
13:52:45 <heyongli> another issue is work partition, any idea?
13:52:51 <baoli> heyongli, the spec doesn't talk about how to implement it.
13:53:16 <baoli> yet.
13:53:19 <heyongli> bali, so my comments is a suggest way
13:53:40 <baoli> So yo are welcome to give more details on how to integrate this module with the existing pci infra
13:53:50 <irenab> baoli: I think some level of implementation details will be required in order to get this spec approved
13:54:24 <baoli> irenab, sure
13:55:05 <heyongli> i would like take patch for whitlist, pci stats, any common and not for sriov only feature.does it ok? baoli seems more likely taken the --nic and vif part, baoli is expert of this
13:56:03 <baoli> heyongli, we should combine the ideas behind your patch and my patch and decide what's the best way moving forward
13:56:33 <irenab> I think it worth that everyone will review the spec especially the proposed changes and once seems that all covered, people may take parts to assign for implementation
13:56:38 <heyongli> my bp ,the extra info things for pci offloading and GPU's enhancement will changed according to this limited mulit tag approach.
13:57:45 <heyongli> baoli, the ideas need be sync, work partition is another thing to consideration.
13:58:32 <baoli> heyongli, agreed
13:59:04 <baoli> heyongli, are you coming to Atlanta?
13:59:21 <heyongli> i'm very sorry, very likely not
13:59:42 <baoli> that's bad.
13:59:47 <heyongli> but yunhong will be there, i will offloading anything to him for Atlanta
14:00:00 <irenab> so plan for next week, is to complete spec as much as possible, try to get John on  next PCI meeting
14:00:16 <baoli> irenab, that sounds good
14:00:24 <baoli> #endmeeting