13:00:12 <baoli> #startmeeting PCI Passthrough 13:00:12 <openstack> Meeting started Tue Feb 25 13:00:12 2014 UTC and is due to finish in 60 minutes. The chair is baoli. Information about MeetBot at http://wiki.debian.org/MeetBot. 13:00:13 <openstack> Useful Commands: #action #agreed #help #info #idea #link #topic #startvote. 13:00:15 <openstack> The meeting name has been set to 'pci_passthrough' 13:00:42 <baoli> hello 13:00:49 <irenab> hi 13:01:16 <sadasu> hello 13:02:17 <baoli> Let's wait a couple of minutes for others to join 13:02:25 <irenab> good news for neutron, all general patches are merged 13:02:40 <irenab> baoli: ok 13:02:56 <sadasu> irenab, your vnic_type changes had merged yesterday thanks! 13:03:23 <sadasu> rkukura's vif_details also got merged yesterday 13:03:24 <irenab> also rkukura's patches are in from today 13:03:33 <irenab> also binding:profile 13:03:38 <sadasu> but I am not sure about binding:profile diffs 13:03:42 <sadasu> ok...cool 13:05:36 <beagles> o/ 13:05:42 <irenab> I need to rework current Mellanox MD to support vnic_type and then back to generic parts for neutron pci_info,.. 13:05:53 <irenab> beagles:hi 13:07:18 <sadasu> markmcclain, hi 13:08:22 <sadasu> my BP for ML2 driver for ucs manager : https://blueprints.launchpad.net/neutron/+spec/ml2-ucs-manager-mechanism-driver 13:08:41 <sadasu> got moved to Juno inspite of posting WIP diffs 13:09:19 <sadasu> I was dependent on 3 BPs that were changing generic portbinding code 13:09:26 <sadasu> and I was dependent on that 13:09:36 <sadasu> they got mergerd yesterday 13:10:10 <sadasu> becuz of this i, can it be brought back to Icehouse? 13:10:50 <irenab> baoli: shall we start the meeting? 13:11:10 <baoli> irenab, yes, let's get started. 13:11:13 <sadasu> markmcclain, you here? 13:11:44 <sadasu> irenab, just hoping to get this clarified from markmcclain 13:12:11 <baoli> sadasu, I saw the comments from markmcclain 13:12:18 <irenab> on current status, my understanding that neutron SRIOV MD parts can be worked out, since all generic parts are in 13:12:41 <baoli> I'd like to share some good news. 13:12:56 <irenab> baoli: go ahead 13:13:03 <sadasu> waiting 13:14:32 <baoli> Last night, I updated my WS and got Irenab's patch, and I manually merged rkukura's profile patch (which just got merged) 13:15:11 <baoli> together with my changes in the nova side, this is what I can do now: 13:15:43 <baoli> tag the whitelist with attributes, for example with "physical_network":"physnet1" 13:16:28 <baoli> create an instance with 'nova boot --nic port-id=' and 'nova boot --nic net-id=, vnic-type=' with SRIOV 13:16:37 <baoli> create an instance with macvtap 13:16:57 <baoli> create an instance with both pci alias (existing functionality) and sriov vnic in the same time 13:17:23 <sadasu> very cool 13:17:32 <baoli> The only thing that's missing in the picture is the vif_details which can only be set by the vendor MD 13:17:41 <sadasu> yes 13:17:48 <irenab> baoli: sounds great 13:17:59 <irenab> is 'nova boot --nic net-id=, vnic-type=' with SRIOV part of your code? 13:18:19 <irenab> did you manage to extend nova api? 13:18:23 <baoli> irenab, yes, I made that change as well 13:19:06 <irenab> baoli: cool. When you will be ready to upload the changes? 13:19:51 <irenab> I think I can try to update the https://review.openstack.org/#/c/74464/ to return vif_details in coming days 13:20:06 <baoli> irenab, i'm planning to do it today or tomorrow. I need to remove some of the changes that enables it work on my testbed 13:20:22 <irenab> so we can try to check it end 2 end 13:20:44 <irenab> baoli:all heyongli's code is upstream? 13:20:59 <irenab> I mean, all the code you need to use 13:21:30 <baoli> irenab, all I'm using is the code that is already merged 13:21:51 <irenab> baoli: good news indeed 13:22:02 <baoli> I will send you guys an email to provide instructions on how to use it 13:22:11 <irenab> waiting for the patch 13:22:17 <irenab> and email 13:22:21 <irenab> baoli: thanks! 13:22:35 <irenab> is there any update on documentation? 13:22:36 <baoli> irenab, my pleasure. 13:22:59 <sadasu> baoli, that is great progess..thanks 13:23:34 <baoli> sadasu, yes indeed. At lease, it would allow us to test individual md end to end 13:24:07 <beagles> great stuff! 13:24:11 <baoli> s/lease/least/ 13:25:23 <baoli> so stay tuned for the email and patch. 13:25:46 <baoli> last meeting, we were talking about the summit 13:26:30 <irenab> and documentation 13:26:57 <baoli> irenab, that's right 13:27:48 <baoli> documentation wise, shall we start with a summary on how we got here? 13:28:16 <irenab> for summit, I think we need to make a list of what is still open or need to be coded to make it on summit agenda 13:28:28 <beagles> oh whoops.. that's my cue 13:29:11 <rkukura> Just joined - thought these meetings were on Thursday. Sorry! 13:29:14 <beagles> I didn't get as much time to work on this last week as I had hoped, so its still ongoing. I found a lot of info I was looking for in the google docs and past emails 13:29:57 <irenab> beagles: I think you better check with us, especially baoli for what is relevant 13:30:21 <beagles> irenab, good point 13:30:48 <irenab> rkukura: thanks for all patches. All are merged! 13:31:07 <beagles> I'll send a summary along with references via email for verification of relevance today 13:31:43 <irenab> beagles: great 13:32:01 <baoli> rkukura, your binding:profile patch works for me. and Irenab's patch works great as well 13:33:14 <rkukura> baoli, irenab: glad to hear it 13:33:19 <baoli> rkukura, just recap. I have something working together with you guy's patches, and I will send an email and update the patch soon. 13:34:10 <baoli> rkukura, we are now talking about documentation and summit 13:34:41 <rkukura> I'll read the log afterwards. Thanks! 13:34:54 <irenab> comming back to summit agenda, my feeling we need to define how the PCI resources are defined and requested for network will look like. It was mainly discussed during the initial PCI meetings, and no consensus was reached 13:35:34 <irenab> host aggreagates, PCI flavor, NIC Flavor,... 13:36:19 <baoli> irenab, that's right. So far, we have explored: pci group, nic flavor, pci flavor, host aggregates. 13:37:46 <baoli> let's put them on the table: pros and cons, use cases, etc 13:39:22 <irenab> baoli: good idea. Can you make some initial table, defining each option. We can update it 13:39:53 <baoli> irenab, sure. I can put it onto the meeting wiki, and people can start updating it. 13:40:23 <irenab> having it on the table, we can work on summit session content and maybe even POC to show 13:40:50 <baoli> irenab, sounds good. 13:41:11 <irenab> Another issue we discussed was neutron aware nova-scheduler 13:41:45 <baoli> irenab, I'm a little bit fussy about that. maybe you can give a description on that first. 13:43:04 <baoli> or you can add a section to the wiki for the idea. 13:43:19 <irenab> it was mainly driven by Ian, but I think rkukura may add more details, since it was also relevant for non SRIOV cases, on how to get VM on Host that has certain physicla networks available 13:43:49 <irenab> baoli: I'll take a look to find Ian's doc on this. 13:44:03 <baoli> irenab, sounds good. 13:45:14 <baoli> We touched upon the non SRIOV cases in past IRC meeting. I remember scheduling is not involved. rkukura can correct me. 13:46:23 <irenab> baoli: found the doc #link https://docs.google.com/document/d/1vadqmurlnlvZ5bv3BlUbFeXRS_wh-dsgi5plSjimWjU/edit 13:46:53 <irenab> there is a section for neutron aware scheduling 13:48:05 <rkukura> As I recall, all I had said on this was that there are other cases where nova's scheduling would need to take neutron into account - making sure there is connectivity, or trying to manage latency, etc. 13:48:32 <baoli> irenab, yes, I read that before. I'll read it again. 13:48:35 <irenab> I guess we need to validat the curent was to request vnic_type and tagging resources, and if its good enough, no additional scheduling enhancments are required 13:49:29 <irenab> svalidate the current way 13:49:59 <baoli> irenab, in any case, we should think about it and see if we should do anything about it. 13:50:23 <irenab> need also work item for Horizon to make it possible to specify vnic_type 13:51:05 <baoli> irenab, we have someone here to work on horizon. I can talk to him. But I think that it wouldn't make icehouse 13:51:43 <irenab> baoli: it can be great 13:53:02 <irenab> baoli, sadasu: question on port_state_up attribute 13:53:41 <irenab> How would you support the change of port_state_up without local neutron agent? 13:54:47 <sadasu> I am stuck there too and exploring diff options 13:55:14 <sadasu> MD will check with the actual device is one option 13:55:24 <irenab> I am adding the agent anyway, need to see if you require one, it can be generic enough... 13:55:29 <sadasu> not sure if that will work or break anything 13:55:47 <baoli> irenab, do you mean the admin-state of the port? 13:56:03 <irenab> baoli: exactly 13:57:05 <irenab> rkukura: Do you know when design summit sessions should be proposed? 13:57:20 <sadasu> I don't think there is an issue with the work item itself 13:57:41 <baoli> before a neutron sriov port is bound to an instance, you can't admin down/up the sriov device, right? 13:57:43 <sadasu> except for this case, we don't need an agent at all 13:57:47 <rkukura> irenab: Not sure 13:57:59 <sadasu> hence my questions to rkukura about this a couple of weeks ago 13:59:22 <baoli> we have a couple of minutes left. I need to move on to another meeting. So let's continue next week. 13:59:36 <irenab> I think we need to reach some level of understanding what is all required work for Juno time frame and prepare it to the summit, to get core's on it from the start. 13:59:50 <irenab> So let's do it next meeting 13:59:58 <baoli> Thanks everyone, and see you guys next week. 14:00:01 <sadasu> irenab, +1 14:00:13 <sadasu> thanks all 14:00:16 <irenab> thanks 14:00:16 <baoli> #endmeeting