16:01:27 <mlavalle> #startmeeting neutron_performance 16:01:28 <openstack> Meeting started Mon Nov 5 16:01:27 2018 UTC and is due to finish in 60 minutes. The chair is mlavalle. Information about MeetBot at http://wiki.debian.org/MeetBot. 16:01:29 <openstack> Useful Commands: #action #agreed #help #info #idea #link #topic #startvote. 16:01:31 <openstack> The meeting name has been set to 'neutron_performance' 16:01:44 <njohnston_> o/ 16:01:59 <rubasov> o/ 16:02:04 <mlavalle> hey guys 16:02:20 <mlavalle> rubasov: when are you arriving to Berlin? 16:02:40 <rubasov> mlavalle: monday early afternoon 16:02:49 <rubasov> mlavalle: the same plane as gibi 16:03:12 <mlavalle> I'll be there since Friday morning. I'm attending the edge computing hackathon 16:03:24 <mlavalle> during the weekend 16:03:42 <rubasov> mlavalle: sounds interesting 16:04:06 <bcafarel> this time it's not my fault holding the others from being there :) 16:04:18 <mlavalle> bcafarel: duly noted 16:04:21 <mlavalle> LOL 16:04:55 <mlavalle> let's wait one more minute 16:04:59 <mlavalle> for slaweq 16:05:59 <mlavalle> ok, let's get started 16:06:10 <mlavalle> #topic last week's action items 16:06:34 <mlavalle> First rubasov to add top request offenders to rally job 16:07:01 <rubasov> I'm sorry to report but I didn't make much progress 16:07:07 <rubasov> started to learn about rally 16:07:15 <rubasov> but then got busy with other stuff 16:07:23 <mlavalle> yeah, I imagine 16:07:28 <mlavalle> the demo, maybe? 16:07:32 <rubasov> yep 16:07:40 <mlavalle> np 16:07:45 <slaweq> hi 16:07:52 <rubasov> but I still have it on my todo list 16:07:55 <slaweq> sorry for being late 16:08:04 <rubasov> so I will get to it at some point 16:08:08 <mlavalle> let's move that action item for the next meeting 16:08:23 <mlavalle> #action rubasov to add top request offenders to rally job 16:09:04 <mlavalle> next one is slaweq to create a experimental rally job with osprofiler enabled 16:09:34 <slaweq> so, I proposed patch for rally https://review.openstack.org/#/c/615340/ 16:09:55 <slaweq> and then I have WIP patch which depends on it: https://review.openstack.org/#/c/615350/ 16:10:12 <slaweq> it basically works, and I can see trace id in rally result 16:10:34 <slaweq> but now I need to copy somehow redis db to job result files 16:10:44 <slaweq> so I will continue work on this in next days 16:11:11 <mlavalle> Great progress slaweq. Thanks 16:12:03 <mlavalle> next action item was mlavalle to simplify the DB traces 16:12:33 <mlavalle> I just proposed this patch https://review.openstack.org/#/c/615585 16:13:31 <mlavalle> I made sure that the code that I am removing is the only one in Neutron that add traces for SQLAlchemy 16:14:07 <mlavalle> so after this one merges, we should get much cleaner reports 16:14:17 <mlavalle> with less duplicate information 16:14:34 <mlavalle> which will make the reports easier to analyze 16:14:41 <mlavalle> ok 16:14:51 <mlavalle> that's it for action items 16:15:00 <mlavalle> #topic On demand agenda 16:15:26 <njohnston_> I just wanted to note that the ipam work I have been doing has some promising speed results 16:15:46 <mlavalle> how so? 16:17:09 <njohnston_> https://review.openstack.org/#/c/592164/ look at my last comment 16:17:41 <njohnston_> the rally task "Action neutron.create_port (x50)" looks like it is ~30% faster 16:18:25 <mlavalle> wow, nice! 16:18:42 <mlavalle> should we add some profiling calls to the IPAM code? 16:18:45 <njohnston_> This is mostly accomplished by taking the code in ipam to try to randomly select new IPs (to avoid contention) and replacing it with random.sample 16:19:33 <njohnston_> mlavalle: We could, although I think we may have wrung all the speed gains we can out of this code. But I'm definitely open to adding profiling to confirm it. :-) 16:20:39 <mlavalle> mhhhh, no, if you don't see it evidently worthwhile at this point, let's wait until we can see results from the rally jobs 16:21:20 <mlavalle> with more data, we may conclude that further investigation is warranted 16:21:41 <mlavalle> but at this point I think we are still "shooting in the dark" 16:22:13 <mlavalle> but I think this work will definitely have impact 16:22:23 <mlavalle> becuase port creation is among the top offenders 16:22:34 <mlavalle> which this work with IPAM impacts 16:23:05 <mlavalle> great job, njohnston_ 16:23:26 <njohnston_> I agree - and I want to thank tidwellr and Roman (I don;t know his IRC) for great review feedback 16:23:33 <njohnston_> would not have happened without them 16:24:06 <slaweq> njohnston_: Roman's nic is gryf :) 16:24:26 <njohnston_> Ah thanks :-) 16:24:50 <mlavalle> yeah, tidwellr did a lot of work in IPAM code a couple of yeards ago 16:25:18 <mlavalle> and gryf was probably with John Bellamaric's group 16:25:28 <mlavalle> and they also did great work in IPAM 16:26:04 <mlavalle> ok, let's move on 16:26:16 <mlavalle> I think that at this point what we need is data 16:26:37 <mlavalle> to start gleaning insight 16:26:54 <mlavalle> and to get to that point we need to finish our action items 16:26:58 <mlavalle> do we agree? 16:27:04 <njohnston_> +1 16:27:10 <slaweq> +1 16:27:27 <mlavalle> so let's focus on concluding our action items 16:27:32 <rubasov> yep 16:27:40 <mlavalle> and we can re-take the discussion next meeting 16:27:49 <mlavalle> ok? 16:27:56 <njohnston_> sounds like a plan! 16:27:59 <slaweq> ok for me :) 16:28:09 <mlavalle> ok, guys, thanks for attending 16:28:25 <rubasov> thank you guys 16:28:34 <njohnston_> thanks all 16:28:42 <mlavalle> #endmeeting