13:02:06 <baoli> #startmeeting PCI Passthrough 13:02:07 <openstack> Meeting started Tue Sep 30 13:02:06 2014 UTC and is due to finish in 60 minutes. The chair is baoli. Information about MeetBot at http://wiki.debian.org/MeetBot. 13:02:08 <openstack> Useful Commands: #action #agreed #help #info #idea #link #topic #startvote. 13:02:10 <openstack> The meeting name has been set to 'pci_passthrough' 13:02:33 <baoli> Hi there 13:02:37 <heyongli> hi 13:02:40 <rpothier> hi 13:02:43 <irenab> hi 13:03:10 <baoli> let's start with bugs 13:03:32 <baoli> https://bugs.launchpad.net/nova/+bug/1370348 13:03:35 <uvirtbot> Launchpad bug 1370348 in nova "Using macvtap vnic_type is not working with vif_type=hw_veb" [Undecided,Confirmed] 13:04:28 <baoli> irenab, is it ok to fix it from the nova side? 13:04:34 <irenab> according to Daniel comment, it should be OK to configure vlan in scope of nova 13:04:51 <irenab> baoli: fine with me. Less work on neutron side :-) 13:05:06 <baoli> irenab, so do we create a vlan interface or just set the vlan in the vf? 13:05:33 <irenab> baoli: set vlan in the vf 13:05:49 <baoli> irenab, ok. is itzik going to work on this? 13:06:15 <irenab> baoli: I do not think so 13:07:06 <baoli> irenab, that's fine. In that case, let me post the review soon. 13:07:26 <irenab> baoli: great, thanks 13:08:08 <baoli> heyongli, how is the resize investigation going? 13:08:45 <heyongli> i checking this problem but pull by other thing 13:08:47 <heyongli> for now 13:09:07 <heyongli> resize from 1 pci to more pci devices is work.. 13:09:07 <baoli> heyongli, ok 13:09:33 <heyongli> but not very carefully check , but at least host not die 13:09:49 <heyongli> and OS response ok, i need to dig more to reproduce the bug 13:09:52 <baoli> heyongli, that's no what I saw. Did you check the resulting domain xml, and did you try to login into the instance? 13:10:29 <heyongli> not now , i land it to our CI, not that detail yet 13:11:07 <baoli> In the xml, the number of PCI devices are not correct, Also, I couldn't login to the instance. 13:12:06 <heyongli> ok baoli , i'm gong to extern the test case . but this week is our vacation wekk 13:12:08 <heyongli> week 13:12:41 <heyongli> i will not response mail 1 week and miss next meeting 13:13:00 <baoli> heyongli, ok, no problem. I'll look into it as well whenever time allows 13:13:12 <heyongli> ok, thanks then 13:14:03 <baoli> any other issues/bugs to discuss? 13:14:43 <irenab> What about next development, accordign to etherpad? 13:15:03 <baoli> irenab, that's the next thing we'll discuss after bugs 13:15:11 <baoli> https://etherpad.openstack.org/p/kilo_sriov_pci_passthrough 13:15:31 <heyongli> irenab, or anyone do you have any design about the next kilo? 13:15:56 <baoli> So once the kilo BP/specs opens up, we should add ours 13:16:11 <sadasu> I have an update... 13:16:33 <sadasu> the Cisco UCS manager plugin now has to support non-sr-iov, regular VMs 13:16:39 <heyongli> baoli, yeah, before that we need a work partion or how to coopration 13:17:12 <baoli> heyongli, yes. 13:17:28 <heyongli> any idea? 13:18:14 <heyongli> or show many resource you guy got? 13:18:19 <irenab> I think we need to see what is important to progress 13:18:21 <baoli> heyongli, how about we pick up one of the first two items, one for each? 13:18:36 <heyongli> ok for me 13:19:10 <baoli> irenab, yes, we need to agree on the priorities first. 13:19:16 <heyongli> which one you have worked more on? 13:19:30 <baoli> heyongli, the first one 13:19:42 <heyongli> plugin/migration seems more important to us 13:19:51 <irenab> what about number 4? 13:20:04 <heyongli> ok, baoli, you work more on the live migration 13:20:32 <baoli> irenab, yes, that's a good feature to have 13:21:18 <irenab> also we probably need to have option to make nova boot to work without previously port create 13:21:45 <irenab> I'll look into number 4 for more details 13:22:02 <heyongli> irenab, we seems lots of problem to altting the nova api 13:22:02 <baoli> irenab, that's going to be a tough battle. But with the default added to the network, it's going to be a given 13:22:09 <baoli> irenab, that's great 13:22:58 <irenab> baoli: I also think that having number 4, will make nova boot work. But still no fullGUI support 13:23:18 <sadasu> let me add that to the etherpad..I don't think it is specifically mentioned 13:23:20 <heyongli> irenab, thanks we can push that by all of us 13:23:43 <heyongli> sadasu, seems not, thanks 13:23:57 <heyongli> full GUI? 13:24:02 <heyongli> what that? 13:24:28 <irenab> heyongli: work via dashboard, without cli 13:24:28 <sadasu> heyongli: whats your question? 13:24:58 <baoli> irenab, forgot to report that one of our engineers will enhance horizon for creating sr-iov ports 13:25:05 <irenab> with number 4 supported, still need to create network with default vnic_type associated, and this will be probably via cli 13:25:06 <sadasu> heyongli: I think some folks from Cisco are starting to look at this from Horizon's pov 13:25:07 <heyongli> irenab, got, thanks 13:25:34 <heyongli> thanks clearify sadasu 13:25:34 <irenab> sadasu: Can you please share more details? 13:26:07 <baoli> heyongli, what's number 14? 13:27:05 <heyongli> baoli, i remember lot's of people don't like whitelist and call for module solution 13:27:11 <heyongli> we can try 13:27:16 <sadasu> irenab: for a minute I thought the discussion was about GUI support for sr-iov config 13:27:44 <irenab> sadasu: I believe it was 13:27:46 <sadasu> irenab: but now, it doesn't seem to be 13:27:52 <baoli> heyongli, can you add more details into the etherpad? 13:28:03 <heyongli> baoli sure 13:28:17 <irenab> please let's have one discussion at a time :-) 13:28:21 <baoli> sadasu, irenab, let's talk about gui then 13:28:36 <sadasu> so for a start we would like to be able to configure the diff vnic-types via Horizon during port-creation step 13:29:23 <irenab> sadsu: so, it would be possible to set vnic_type per chosen network for VM? 13:29:50 <sadasu> as a stretch goal, they are also looking into displaying available PCI devices per compute host 13:30:21 <baoli> irenab, it's the ability to specify vnic-types for a port in horizon 13:30:35 <sadasu> irenab: that is correct 13:30:44 <baoli> irenab, if you added number 4, then that should be added into horizon as well 13:30:53 <baoli> s/added/add 13:31:17 <irenab> given that, do we still want default vnic_type per network? 13:31:30 <irenab> baoli: got it 13:31:43 <sadasu> there is no API that currently provides available PCI devices per compute host that Horizon can readily leaverage 13:32:09 <irenab> which takes me to number 6 and 7 on etherpad 13:32:41 <heyongli> sadasu, the api might there, i remember a extention to the host is added, but i need check it with our guy. 13:33:01 <sadasu> can we also add this API to list of Kilo items on the etherpad? 13:33:12 <baoli> heyongli, it doesn't seem to be merged. 13:33:25 <irenab> sadsu: I think we should 13:33:29 <sadasu> heyongli, can you confirm by next week's meeting? 13:33:38 <heyongli> baoli, i need check that, 13:33:53 <baoli> well, I think I'm wrong. 13:33:56 <heyongli> and fill the table after i got detail 13:34:10 <baoli> check 'nova help' | grep pci 13:34:28 <heyongli> sadasu, next meeting i will not attend, cause of national day of China 13:34:45 <sadasu> but keep in mind that even if API is present / will be available soon, we haven't still looked into Horizon specific issues 13:35:13 <irenab> Does anyone have ideas what we desired for admin/tenant APIs? 13:35:22 <sadasu> heyongli: will you be able to confirm the current status of those changes via email to the ML sometime during the week then? 13:35:34 <heyongli> #AR baoli, there might some story there , i will check that any way. about the api ext about show host 13:35:54 <heyongli> sadasu, after 1 week, i can ,yes 13:36:04 <sadasu> irenab: can u be more specific? 13:36:21 <sadasu> heyongli: thanks...that will do 13:36:33 <heyongli> that APi not a stand alone one 13:36:45 <heyongli> might be show hyper visor thing, i fogot detail. 13:36:58 <irenab> sadasu: you mentioned 'get available pci devices'. What else should be available for admin/tenant? 13:37:23 <baoli> heyongli, I think it's not merged. 13:37:31 <heyongli> item 8, 13:38:08 <heyongli> baoli, i remember because of API version things, maybe. if not, we will restarting that job i think 13:38:08 <irenab> sadsu: via API/ flavor 13:38:34 <baoli> heyongli, thanks 13:38:48 <sadasu> irenab: just have that one so far..and should it be admin only API? 13:38:50 <heyongli> item 8, i had something enhancement about schedule filter also, might could merge the this worklist 13:39:12 <irenab> Shall we go for more advanced requirements, like BW 13:39:32 <baoli> heyongli, do you have a patch or spec available? 13:40:03 <heyongli> baoli, i had one, will paste some detail after vacation 13:40:14 <baoli> heyongli, sounds good 13:41:31 <heyongli> item12, i will check that bug, i think something been addressed and some not. i will figure it out 13:42:36 <baoli> heyongli, yes, the sr-iov work that's merged have some of them addressed already 13:42:58 <irenab> sadasu: regarding GUI support, is there already some patch? 13:44:33 <sadasu> irenab: no...not even close...just started looking into it yesterday 13:44:56 <irenab> sadasu: thanks 13:45:19 <sadasu> more importantly to see if, any action items have to be added on Nova/Neutron side to support Horizon changes 13:45:34 <baoli> irenab, sadasu, how about the MD refactoring 13:45:35 <irenab> sadasu: agree 13:46:44 <sadasu> baoli: wanted to discuss that if discussion for nova items are done 13:46:50 <irenab> sadasu mentined some new requirements for cisco MD 13:46:51 <heyongli> baoli, irenab, refracting is good, intresting to me 13:47:24 <sadasu> so the requirements for Cisco UCS manager MD now include support of non-sr-iov VMs 13:47:38 <baoli> I think that we need to dedicate some time on it in next meeting 13:47:46 <irenab> on Mellanox side, there is also InfiniBand support that I need to take care of 13:48:11 <sadasu> and my feeling that this MD is going to keep evolving to support other non-sr-iov features over the next few releases 13:48:38 <irenab> baoli: OK. I suggest we exchange some emails regarding the requirement prior the meeting, to make it more efficient 13:48:49 <sadasu> so, keeping both of that in mind, how can we proceed with the refactoring? 13:49:00 <baoli> irenab, let's do that, can you guys start the thread? 13:49:16 <sadasu> we have 10 mins...lets do some now 13:50:01 <sadasu> Kilo: Mellanox : InfiniBand, Cisco : Non-sr-iov 13:50:31 <baoli> sadasu, let's agree on the priories if we can. 13:50:49 <sadasu> irenab: are you planning to add Infiniband suppport to the existing MD? 13:51:21 <sadasu> baoli: in cisco's case, kilo support of sr-iov is not optional 13:51:26 <irenab> sadsu: still not decided. There is Mellanox MD that may evolve to support it 13:51:49 <sadasu> I am guessing it is the same with mlnx 13:51:59 <irenab> sadasu: Will try to figure it out till next meeting 13:52:12 <irenab> sadsu: yes 13:52:28 <sadasu> irenab: ok 13:52:55 <baoli> sadasu, irenab, let's think it through offline, and start the thread. 13:53:07 <baoli> ok, back to priorities 13:53:10 <irenab> baoli: fine 13:53:19 <sadasu> I am wondering if we our respective neutron MDs are diverging significantly in kilo 13:53:28 <sadasu> baoli: sure 13:54:54 <baoli> so 1, 2, 13 at the top of list? 13:54:56 <irenab> The question is if we want to have code reuse for both MDs 13:55:11 <irenab> baoli: and 4? 13:55:33 <baoli> irenab, yes 13:55:44 <irenab> somehow numbering is wrong now... 13:56:35 <baoli> irenab, sadasu, refactoring is important as well. If it's really something desirable, then it should be addressed in kilo. 13:57:08 <irenab> baoli: we should define what we want to achive with refactoring 13:57:27 <irenab> code reuse/ singe MD with drivers,... 13:57:31 <sadasu> baoli: I consider that as Neutron priority list item 13:57:48 <sadasu> baoli: let us figure of Nova priorities first 13:57:59 <baoli> so nova side, 1, 2, 13, neutron side, 4, 10 13:58:28 <baoli> irenab, I agree. 13:59:02 <irenab> good. Let's discuss it during our next meeting 13:59:14 <baoli> well, time is almost up. So we still have a little time to think about. We should start the BPs/Specs, soon so that we can make it for Kilo 13:59:30 <sadasu> baoli: +1 for priorities listed above 13:59:58 <irenab> it feels like something that we do not need session at the summit, agree? 14:00:45 <baoli> thanks everyone. Has to end this meeting now 14:00:47 <baoli> #endmeeting