21:00:31 #startmeeting scientific-sig 21:00:32 Meeting started Tue Nov 12 21:00:31 2019 UTC and is due to finish in 60 minutes. The chair is oneswig. Information about MeetBot at http://wiki.debian.org/MeetBot. 21:00:33 Useful Commands: #action #agreed #help #info #idea #link #topic #startvote. 21:00:35 The meeting name has been set to 'scientific_sig' 21:00:42 greetings! 21:00:43 woohoo! 21:00:47 hello! 21:00:52 (rumble rumble) 21:01:42 #link Agenda for today https://wiki.openstack.org/wiki/Scientific_SIG#IRC_Meeting_November_12th_2019 21:01:45 #chair martial 21:01:46 Current chairs: martial oneswig 21:01:47 hey martial 21:02:07 How's things y'all? 21:02:29 things are good, busy :) 21:02:45 and getting ready SC19-rumble too ! 21:03:02 awesome. What day is the BoF again? 21:03:04 doing alright. busy. trying to the flu my son has attempted to bring back from daycare a second time ;) SC19 ahead! 21:03:13 *to avoid* 21:03:20 Tuesday? 21:03:40 *I'm prepared, honestly* 21:03:56 Wed 5:15pm 21:04:12 so, Stig, back from China ... how was it? 21:04:28 It was an interesting one. Quite a clash of culturs. 21:04:59 Some oddly officious stuff - to get into the conference each day, passports and bag x-ray, like going through an airport. 21:05:39 The dual-language schedule worked sometimes, people made effort to make it work. 21:06:45 dual-language? 21:07:02 oh wow did not realize it was going to be that different 21:07:22 yeah, i didn’t hear about that 21:07:24 how’d that work? 21:07:28 One of the Chinese language keynote speakers appeared to have an interesting take on the benefits of open source to the Chinese nation 21:07:49 such as? 21:07:57 The plenary sessions had translators. Everything else, you made a guess about what language it was going to be in. 21:07:59 or should I not ask? 21:08:40 There was a talk in chinese that I couldn't follow and wish I could have, on Tik-Tok (the video sharing site), and how they are using mlnx offloads. 21:09:52 Did you see the graphic Marco Passerini shared on the leading vendors for OpenStack deployments? 21:11:01 martial: live translation worked well enough. Some amusing mis-translations (eg "bare metal" -> "naked machines") 21:11:05 what was the link? 21:11:17 funny indeed:) 21:11:25 lol 21:11:34 It was on slack earlier, don't have it to hand 21:12:07 Overall numbers were about the same as Denver I think 21:13:00 I think there was potential for much higher numbers (OpenStack days in Chinese cities are bigger than that), it'll be an interesting discussion point in the wash-up from this summit 21:14:00 seems a little sad given how many of the usual crowd was unable to attend 21:14:05 where is the next one? 21:14:22 There was a frequent theme in the talks, where one presenter would say "my colleagues couldn't join me because..." 21:14:39 Next event's Vancouver again, but a tech summit without marketplace. 21:14:53 oh wow, Vancouver again 21:14:59 Vancouver is a good one 21:14:59 I think then Europe again around this time next year 21:15:12 Wait, did the next summit get announced? 21:15:28 sorry, almost missed the meeting due to daylight savings change 21:15:35 Kind of. I missed the announcement because the keynotes overran and I wanted to go to some forum sessions. 21:15:41 Hi jmlowe, no worries. 21:16:02 Looking forward to seeing a number of you next week 21:16:24 I haven't seen the plan for next year online anywhere 21:17:02 jmlowe: me too. Didn't have a ticket for a bit earlier this week, which was... interesting. I think I never got the mails from spargo with the ticket details. 21:17:07 I have it now. 21:17:30 martial: do you use Grafana? 21:17:44 Is it part of a Data Machines product? 21:18:24 it is part of the tool set we have indeed 21:19:24 also prometheus 21:19:25 There was some concern in the corridors about changes to the t&cs here https://grafana.com/trademark-policy/ which mean products incorporating grafana now require the written permission of grafana labs. 21:19:54 oh weird 21:20:19 indeed, weird. We don't make a product so the details somewhat passed me by. 21:20:22 how do you define "incorporating"? ... started as part of a docker-compose? 21:21:02 I would have thought more generally than that. If it's installed as part of something you sell. 21:21:46 we do not sell products per say so still kind of hard to "grasp" but thanks will pass the info along 21:21:51 was Blair there with you? 21:22:25 No, not this time. Sam Morrison and Andy Botting were holding the fort for the antipodeans, did a great job. 21:24:25 The major disappointment for the SIG was the scheduling of SIG time. It was a table in the PTG, which meant there weren't many operators around. But also, there were almost no locals in the PTG. And there wasn't a cross-project time when (eg) the Ironic cores could come along and join in. 21:24:52 As a result, there were ~4 of us compared with ~100 on a good summit. 21:26:47 It's so much easier for a number of us to go to, well, anywhere but China 21:27:22 jmlowe: yeah, that was a running theme. 21:27:51 heh 21:28:10 still a bummer 21:29:00 Hi folks 21:29:06 I hope the foundation managed to stir up some interest in the open source community among the locals, but it did look like a cultural divide. 21:29:18 hi trandles, just in time :-) 21:29:52 Sorry, home sick today and lost track of the time 21:30:07 get well soon, you've got a flight to catch 21:30:28 Lol yeah...started on Friday, yesterday was rough but I'm over the hump I think. Should be in the office tomorrow. 21:31:06 Reading back...I haven't seen any formal announcement on upcoming summits, but Vancouver would be good. 21:31:46 Did anyone (other than Tim Bell) notice that neutron is talking about deprecating linuxbridge support? I know some of this community, including CERN, was using it everywhere. 21:32:01 yes I saw that and am surprised by it. 21:32:21 I think it's still the simplest and fastest of the software bridges. 21:32:44 Definitely the simplest. Has OVS performance ever caught up? 21:32:49 No. 21:32:57 At least, it hadn't last week when we last tried it. 21:33:09 This surprises me. It’s definitely the simplest. 21:33:22 Actually, correction, that was OVS vs Tungsten Fabric 21:33:26 hrm...maybe this SIG needs to speak up then 21:33:39 jmlowe: you still using it at large scale? 21:33:49 Though OVS had a decent set of tools for debugging, etc that was nice 21:34:11 I also picked up a pre SC cold on Friday 21:34:16 One of the reasons they cite is feature set falling behind other solutions. Conversely, the simplicity is nice when you care about raw performance. 21:34:47 I'm not happy about this linuxbridge deprecation 21:34:54 trandles: +1 for me 21:35:12 I'm still using linuxbridge at least through 2020 21:35:29 I think the issue is nobody's stepping in to look after it with the Neutron project. 21:35:53 Those people are I expect all paid to work on shiny stuff. 21:37:30 oneswig : ever test the performance of ovs vs ovn? 21:37:42 Funny you should ask. 21:38:10 I benchmarked it for Sydney - 2 years ago - using 100G CX5 NICs 21:38:17 1 sec 21:39:28 for bonus points ever try the ovs offload in connect-x cards? 21:40:10 yes it's brilliant, like flossing with dynamite 21:40:20 occasionally, it blows up :-) 21:40:56 oh well the advantage of having disposable hardware :) 21:41:17 Just uploading the presentation somewhere. 21:41:36 oneswig, saw the foundation announcement, congrats 21:42:48 trandles: thanks :-) 21:42:49 ? 21:42:55 i think i missed the memo 21:42:55 Here's the slides http://www.stackhpc.com/resources/5-ways-with-hypervisor-networking.pdf 21:43:26 rbudden: we got ourselves on the marketplace and joined the foundation, that's all :-) 21:43:35 oh nice :) 21:43:39 on the upside I did see that sriov live migration is a thing in train 21:44:15 jmlowe: I believe so, with Netronome smart NICs 21:44:52 bah, I didn't see that qualifier 21:44:57 On that presentation linked above, there's OVN vs OVS vs LinuxBridge vs SR-IOV vs ASAP2 - look for a slide titled "TPC/IP performance - iperf" 21:45:40 NB We tested OVN again last week and it was better than this but still not the quickest thing out there. 21:49:07 Actually, I hear the hardware OVS offload is a good deal more mature now and in production at large scale. 21:51:09 looks like ASAP2 might be a real contender 21:53:27 I think so, potentially. It would be good to see how performance degrades with increasing numbers of flows but at any level it will beat a software switch by a huge margin. 21:53:49 We should talk SC 21:54:11 Any good schedule picks? 21:54:47 Well there's this workshop Monday morning 21:55:15 which one is that? 21:55:28 Monday afternoon is CANOPIE 21:56:09 on Sunday I will be at "The Next Platform" event 21:56:50 ah bummer, Monday is also the "NVIDIA 2019 Special Address" 21:57:43 I'll be keeping an eye out for AMD GPU performance comparisons. 21:58:45 Where's the best barbecue? 21:59:52 Nearly time to close, alas 21:59:55 no idea, but we know the Ramen spot if you’re interested ;) 22:00:28 I'll be at Sams #3, the diner 22:00:38 OK y'all time to stop. 22:00:49 #endmeeting