16:01:34 #startmeeting networking_ml2 16:01:34 Meeting started Wed Feb 3 16:01:34 2016 UTC and is due to finish in 60 minutes. The chair is Sukhdev. Information about MeetBot at http://wiki.debian.org/MeetBot. 16:01:35 Useful Commands: #action #agreed #help #info #idea #link #topic #startvote. 16:01:37 The meeting name has been set to 'networking_ml2' 16:02:08 #topic: Agenda 16:02:15 #link: https://wiki.openstack.org/wiki/Meetings/ML2#Meeting_February_3.2C_2016 16:02:32 #topic: Announcements 16:02:49 Anybody has any announcement to make? 16:03:24 Summit CFP deadline was a few hours ago - anyone know when voting starts? 16:04:08 I do not know when the voting starts - 16:04:38 probably in a week or so 16:05:23 If you have submission where you like us to vote, lets know 16:05:44 Lets dive into the agenda 16:05:54 I submitted one called “Understanding ML2 Port Binding” that may be of interest to this group, so keep an eye out for it! 16:06:35 rkukura : very appropriate topic - 16:06:56 rkukura, sure - this is also something I'm interested in :) 16:07:01 actually many people do not understand it 16:07:16 so, it will be a good topic to cover 16:07:37 it also needs to be documented, so I’m hoping the material for the talk can be turned into a start of a document on it 16:08:44 I think you should prepare the material anyway - regardless of the talk being approved or not - it will be a great service :-):-) 16:09:23 #topic: limited-portsec implementation 16:09:32 https://review.openstack.org/#/c/250036/ 16:09:48 I see this is getting good attention 16:10:08 do we need to discuss anything here? 16:11:25 yalei is not here 16:12:15 based upon the number of comments on the patch, it seems to be in good shape = 16:12:41 if anybody has anything else it is better to put it in the review itself 16:13:05 #topic: Modular L2 Agent update 16:13:20 scheuran__ : want to chime in? 16:13:26 sure 16:13:33 the major patchset merged last week 16:13:45 congratulations!!! 16:13:49 now there is the one missing that moves the code into a separate file... 16:14:26 regarding macvtap integration I pushed the first patchsets 16:14:58 can you post the link? 16:14:58 but currently there is a discussion on the rfe going on regarding anti spoofing support in macvtap 16:15:13 #link https://review.openstack.org/#/q/topic:macvtap_agent 16:15:38 seems like this support is very restricted - so we need to understand the details first... 16:16:26 makes sense 16:16:28 #link https://bugs.launchpad.net/neutron/+bug/1480979 16:16:30 Launchpad bug 1480979 in neutron "[RFE] Adding macvtap ml2 driver and agent" [Wishlist,In progress] - Assigned to Andreas Scheuring (andreas-scheuring) 16:16:41 that's it :) 16:17:33 scheuran__ : hats off to you - that is some good work 16:18:05 thanks! I hope it will continue ;) 16:18:43 anybody has any questions/comments for scheuran__ ? 16:20:32 #topic: Routed networks 16:20:47 #link: https://review.openstack.org/#/c/225384/ 16:20:53 hi 16:21:10 neiljerram: hi 16:21:32 So, I reviewed this patch and posted some comments on this 16:22:49 I still owe a review on this 16:22:49 this patch addresses use case for larger deployments - 16:23:21 rkukura : you should - this will tie into HPB 16:23:25 I'm afraid I'm still behind on detailed review of the latest draft. Sorry for that. It means I haven't seen your comments yet. 16:23:49 HPB? 16:24:16 Oh, just worked it out. 16:24:25 (Hierarchical port binding) 16:24:40 right 16:24:42 yup 16:25:28 Well it sounds like we should continue in Gerrit. Your input is very much appreciated. 16:25:52 neiljerram: Is this still targetted at Mitaka? 16:25:58 In a nut shell it provides similar functionality as HPB, but, by using L3 underlay instead of overlay (e.g. VxLAN) networks 16:26:31 rkukura, formally I'm not sure what the current metadata says. But realistically I think that unfeasible 16:27:22 Sukhdev: They are different ways to scale, but I’m not sure its really similar to HPB. We do need to make sure they don’t interfere with each other 16:28:06 rkukura : correct - my comments are related to that - to make sure they do not break HPB 16:28:30 rkukura, On the other hand, I think Carl might say that he'd really like the _spec_ to be agreed by Mitaka. Otherwise the discussion might continue for ever. 16:29:00 rkukura, But I very much doubt that the code will be in place by Mitaka. 16:29:08 I will try to review this ASAP. Agree getting the spec approved now makes sense, even if the implementation won’t make M 16:30:15 This concept was discussed in one of the neutron mid-cycles - we did extensive white-board discussion on this topic 16:31:04 few vendors who want this implemented were present in the room to describe the problem/use-case 16:31:25 this is essentially to address the scale 16:31:32 works very much like HPB - 16:31:46 In fact, I presented HPB in the same meeting - 16:32:40 they want to be able to dynamically build L2 segments and inter-connect them via L3 routers 16:33:13 as oppose to HPB where one would inter-connect them by using overlay network 16:33:39 instead of using L2 overlay on top of L3, they directly want to use L3 16:34:10 this allows them flexibility in IP address assignments 16:35:49 similar to HPB, when nova schedules a VM on a hypervisor, it will assign the host_id and will ignore the IP address for the port 16:36:21 later neutron will bind the L2 segment to the host and assign a segment and IP address 16:37:12 Sukhdev: Does the spec describe how this deferred IP assignment will be implemented? 16:37:40 I think some of the information is there - but, this part still needs to be worked out 16:37:50 there is lots of TODO 16:39:41 the idea is they want to be able to assign the IP address to the segments as they are build (based upon the placements) as oppose to statically at the time of subnet creation 16:40:32 this does change some of basic constructs of neutron 16:41:05 rkukura : In one of my comments, I even mentioned your name - i.e. check with rkurura :-):-) 16:41:14 so, hopefully, they will pull you in :-) 16:41:26 The most significant is that networks are no longer L2 broadcast domains, right? 16:41:50 correct 16:42:12 I think we can have more productive discussion about it once everybody has reviewed the spec 16:42:24 Sukhdev: agreed 16:42:47 neiljerram : would you like to add anything? 16:43:10 So HPB is a way to scale while preserving L2 adjacency for networks, while routed networks is a way to scale while giving that up 16:43:32 rkukura : yes 16:43:41 nothing at this time, thanks. I will look at your comments on the review. 16:44:32 in fact, there is a new knob being added called l2_adjacency which will be set to true/false to support old/new model 16:45:50 sounds good 16:46:01 anything else on this? 16:46:22 #topic: Open Discussion 16:47:47 any body wants to discuss anything? 16:48:41 silence means we are done :-):-) 16:49:07 Thanks folks - it was a good discussion 16:49:18 will see you same place next week 16:49:26 have a great rest of the week 16:49:32 bye 16:49:34 thanks Sukhdev! 16:49:34 bye 16:49:38 #endmeeting