09:00:49 <oneswig> #startmeeting scientific-wg
09:00:50 <openstack> Meeting started Wed Aug 31 09:00:49 2016 UTC and is due to finish in 60 minutes.  The chair is oneswig. Information about MeetBot at http://wiki.debian.org/MeetBot.
09:00:51 <openstack> Useful Commands: #action #agreed #help #info #idea #link #topic #startvote.
09:00:53 <openstack> The meeting name has been set to 'scientific_wg'
09:01:04 <oneswig> Greetings all
09:01:05 <dariov> hello there!
09:01:10 <senk> Hello o/
09:01:23 <priteau> Hi o/
09:01:29 <oneswig> #link today's agenda (such as it is) is here: https://wiki.openstack.org/wiki/Scientific_working_group
09:02:15 <oneswig> Blair has been in touch, he's just returning from holidays, might be able to join us later
09:02:42 <oneswig> Before we begin, any items to add?
09:03:26 <oneswig> I had one - the CFP for HPCloudSys 2016 in Luxembourg closes this week
09:03:49 <oneswig> #link HPCloudSys2016 is here http://2016cloudcom.ux.uis.no/conf/workshops/hpcloudsys.html
09:04:36 <oneswig> OK #topic Summit Scientific OpenStack guide
09:04:50 <oneswig> Flanders from the Foundation was in touch with an outreachy idea
09:05:18 <oneswig> To scan the schedule and pick out some topics WG members and scientific compute uses in general might be interested in
09:05:49 <priteau> That's a very good idea! I haven't gone through the schedule yet
09:05:52 <oneswig> Then generate a SuperUser article and/or mailing of those
09:06:00 <oneswig> priteau: me neither!
09:06:32 <oneswig> #link https://www.openstack.org/summit/austin-2016/categories/ - schedule's somewhere off this page
09:07:09 <oneswig> In the course of this meeting, can we browse and pick out items of interest
09:07:57 <priteau> the URL says "austin", but changing it to barcelona only changes the header
09:08:03 <priteau> the rest of the content is the same
09:08:42 <oneswig> Good grief, you're right!
09:08:45 <oneswig> ooops
09:08:49 <priteau> #link https://www.openstack.org/summit/barcelona-2016/summit-schedule/ Barcelona summit schedule
09:09:47 <oneswig> #link GlusterFS in OpenStack "In this presentation we will talk about the journey we are still on in finding the best fit for a file system inside a large fully-orchestrated cloud application stack" https://www.openstack.org/summit/barcelona-2016/summit-schedule/events/15987/clustered-file-systems-in-your-application-stack
09:11:42 <dariov> wow, that looks interesting to us
09:11:50 <dariov> I’ll tell our people to drop by :-)
09:12:01 <oneswig> #link We have an interest in using Monasca for high performance telemetry, might like to sit in on this one https://www.openstack.org/summit/barcelona-2016/summit-schedule/events/15166/monasca-one-year-later
09:14:02 <oneswig> If anyone sees a talk they'd find interesting for research computing, please chip in!
09:15:00 <priteau> #link https://www.openstack.org/summit/barcelona-2016/summit-schedule/full/ Full Barcelona summit schedule on a single page
09:15:01 <oneswig> #link realistically I think I'll learn a lot from "BrokenStack: failure stories" https://www.openstack.org/summit/barcelona-2016/summit-schedule/events/15448/brokenstack-openstack-failure-stories
09:16:12 <dariov> lol
09:16:30 <priteau> there are 5 talks with HPC in the title
09:16:51 <oneswig> #link A talk from CERN is always interesting and networking is something they've historically struggled with, it'll be interesting to see their solution https://www.openstack.org/summit/barcelona-2016/summit-schedule/events/15865/neutron-at-cern-moving-thousands-of-production-nodes-from-nova-network
09:17:14 <oneswig> priteau: any picks from those?
09:18:01 <priteau> #link Image is Everything: Dynamic HPC VM Repositories using Murano https://www.openstack.org/summit/barcelona-2016/summit-schedule/events/16161
09:18:29 <priteau> #link Neutron and SLURM: Software-defined Networking for HPC Clusters https://www.openstack.org/summit/barcelona-2016/summit-schedule/events/15454
09:18:53 <priteau> #link “Spartan”, a HPC - Cloud Hybrid: Delivering Performance and Flexibility https://www.openstack.org/summit/barcelona-2016/summit-schedule/events/16540
09:18:57 <priteau> all of them really :-)
09:19:07 <oneswig> Nice!
09:19:27 <priteau> blair's talk too of course
09:20:22 <oneswig> #link Blair's talk on Lustre https://www.openstack.org/summit/barcelona-2016/summit-schedule/events/16548/lustre-integration-for-hpc-on-openstack-at-cambridge-and-monash
09:20:40 <oneswig> Any more suggestions?
09:20:44 <priteau> #link HPC, Unikernels and OpenStack https://www.openstack.org/summit/barcelona-2016/summit-schedule/events/15879
09:21:08 <oneswig> priteau: that's one I'm particularly interested in, thanks for reminding!
09:21:39 <oneswig> OK, keep browsing, lets move the topic on
09:21:56 <oneswig> #topic OpenStack / HPC white papers
09:22:17 <oneswig> So over the summer I've been gathering research on five white paper topics on OpenStack and HPC
09:22:34 <oneswig> Three are now in draft form (phew), one's half way and one's not started yet.
09:23:14 <oneswig> What I'm looking for is subject matter experts on the various topics
09:23:33 <oneswig> The two remaining are bare metal infrastructure and high performance data
09:23:53 <oneswig> priteau: you wouldn't happen to know anyone who works on a bare metal infrastructure system would you ? :-)
09:25:06 <priteau> oneswig: I contacted one of our Chameleon users who did a performance comparison of bare-metal vs KVM vs Docker, but no answer :(
09:25:40 <oneswig> priteau: that's OK I think I have enough on overhead, including your previous introductions.
09:25:52 <oneswig> #link HPC and the overhead of virtualisation https://docs.google.com/document/d/1xJJKPltk-ScA4uKrPVnIAHQRptzgViJNMq1kNLWAQTE/edit?usp=sharing
09:25:52 <priteau> I would be happy to be a reviewer on the bare-metal topic
09:26:21 <oneswig> #link OpenStack and HPC network fabrics https://docs.google.com/document/d/1d2nOxPQPARNhmnc0hXBIMW9c_bjDyaUkUZAzMBNPfEA/edit?usp=sharing
09:26:43 <oneswig> #link OpenStack and HPC workload management https://docs.google.com/document/d/1dbN6HRWQuuETqPIw5qYM8Ya6f3RD1WNbGr_hg2CyHPw/edit?usp=sharing
09:27:05 <dariov> I’m afraid we can’t help you on this, guys
09:27:09 <oneswig> priteau: thanks, appreciate that.  Can we make Chameleon into a case study on that?
09:27:10 * dariov feeling a little bit useless
09:28:13 <oneswig> dariov: the articles I'm hoping are useful for people who know about HPC but want to learn more about how it fits into OpenStack. That's the target audience
09:28:20 <priteau> oneswig: If Chameleon fits your requirements for this white paper, I would be happy to!
09:28:42 <oneswig> priteau: Great, thanks.  I'll get my act together and be in touch...
09:29:12 <priteau> I wasn't sure if the testbed itself was a good fit (rather than the research users do with), since it is a very specific use case
09:29:26 <priteau> I guess it will be clearer once I have read your drafts
09:29:57 <oneswig> priteau: Well it is but what I've been looking for is some diversity between the data points, and you're not Bridges and you're not Aurora so that's even better :-)
09:30:06 <dariov> oneswig, then I’m your perfect guinea pig
09:30:36 <oneswig> dariov: any input is gratefully appreciated, thank you
09:31:15 <oneswig> Move on?
09:31:51 <oneswig> #topic Working group activities at Barcelona
09:32:29 <oneswig> Somebody last week suggested the WG meeting might work well with a poster session.
09:32:47 <oneswig> I think this is a good idea, especially if we get a good number
09:32:59 <oneswig> The concern there being it's a bit late to put something together
09:33:07 <oneswig> Any thoughts on this?
09:34:36 <oneswig> Perhaps it is one to consider for a future summit to enable some forward planning
09:35:38 <priteau> How many people would a call for posters reach?
09:36:17 <oneswig> priteau: I am not sure.  I have ~200 mails in my address book, could spam them (and annoy people)
09:36:26 <oneswig> I'd prefer not to
09:36:57 <priteau> We could try academic mailing lists as well. But you're right that is it quite late notice
09:37:27 <oneswig> It works for people who have posters from other conferences
09:37:59 <oneswig> Another idea is to do something like the brown bag talks - lightning talks from WG members
09:38:38 <oneswig> An ideal situation would be to have two adjoining rooms, some refreshments and a mix of the two formats
09:39:51 <priteau> lightning talks may be easier to organize and get people to present
09:40:05 <oneswig> I'll see what can be arranged - space is tight in the venue.  They might even prefer two smaller rooms to one big one.
09:40:15 <oneswig> Thanks priteau good point
09:41:13 <oneswig> After the success in Austin there's also discussion of a WG social one evening.  Probably the Thursday when I understand nothing official is planned
09:42:00 <oneswig> Does anyone know Barcelona at all and can suggest a venue?
09:42:34 <oneswig> A friend of mine, formerly at Barcelona Supercomputer Center, suggested a few places they sometimes used
09:42:48 <priteau> sorry, I have never been to Barcelona
09:44:42 <oneswig> #link cute website, doesn't look vegetarian friendly http://www.asadordearanda.net
09:45:35 <oneswig> Was there anything further on summit activities?
09:45:58 <oneswig> #topic Any Other Business
09:46:14 <oneswig> What has been missed?
09:46:33 <priteau> oneswig: I wanted to ask, do you have any more information about the OpenStack HPC meetup you mentioned the other day?
09:46:49 <priteau> Is it part of http://www.meetup.com/Openstack-London/
09:47:07 <oneswig> The meetup in London, I followed up on it a few days ago and they couldn't get a venue so it has been postponed
09:47:09 <oneswig> Alas
09:47:24 <oneswig> thanks, I'd forgotten about that
09:47:33 <priteau> Removing it from my calendar
09:47:56 <b1airo> hi folks, finally made it
09:48:03 <b1airo> bit of a marathon this evening
09:48:05 <oneswig> Are you going to OpenStack Day UK?
09:48:10 <oneswig> Hi b1airo, welcome back?
09:48:20 <oneswig> (I mean !)
09:48:28 <b1airo> thanks oneswig :-)
09:49:06 <oneswig> We were just on AOB - I think we are pretty much done for the week
09:49:11 <dariov> oneswig, is the “Cambridge meetup” thing still in the pipe?
09:49:34 <priteau> oneswig: I would like to go but I need to check if I can. Is the speaker line-up complete?
09:49:35 <b1airo> any big news i missed discussing?
09:49:47 <oneswig> dariov: Good point, we had to defer over the summer on account of all those bl**dy holidays, lets pick up again
09:49:48 <dariov> (I’m off for 3 weeks now, but something is October should be good)
09:49:57 <oneswig> dariov: b*gger, it rolls on
09:50:03 <oneswig> October it is
09:50:13 <b1airo> i haven't yet reviewed the full summit agenda but looking forward to seeing what you all thought was good
09:50:16 <dariov> yup, my boss here is keen in hosting, if needed
09:50:24 <oneswig> b1airo: We need links into the schedule to recommend
09:50:39 <oneswig> dariov: great, I love a field trip
09:50:52 <dariov> oneswig, cool
09:51:10 <dariov> I’ll be “back in the office” on the 20th of September
09:51:16 <b1airo> ok, i'll try to make sure i get to looking at it soon before work people realise i'm back from leave
09:51:50 <dariov> we can try to find a date in october
09:51:56 <b1airo> else it'll end up being the usual JITSS (just in time summit schedule)
09:51:57 <oneswig> dariov: when are you finishing before your holiday?  Might be good to get something in advance
09:52:06 <dariov> possibly after the 10th, we have a deadline on that day
09:52:27 <oneswig> b1airo: I usually find the events I'm most interested in slightly after they've finished
09:52:30 <dariov> in until friday, but I should be able to respond to email anyway, just slowish
09:52:44 <oneswig> dariov: OK lets follow up offline
09:53:13 <dariov> oneswig, ok
09:53:50 <oneswig> Anything else to raise?
09:54:04 <oneswig> Oh, I had one item
09:54:20 <oneswig> What is considered to be decent TCP bandwidth over VXLAN?
09:54:40 <b1airo> ah, still having issues there?
09:55:01 <oneswig> We have 50G Ethernet NICs.  With SR-IOV we get > 20GBits/s.  With VXLAN, at best 1.6 GBits/s
09:55:32 <oneswig> Still having issues, going for a webex later
09:56:06 <oneswig> But I wanted to find out how many orders of magnitude out we are...
09:56:56 <priteau> oneswig: I suppose that you've checked the MTU size?
09:57:00 <b1airo> sounds like you're 10+ times slower than expected
09:57:29 <oneswig> It's standard MTU in this case, 1400 bytes after encapsulation.  Not sure we can realistically change that for these networks
09:57:53 <priteau> but is the VM configured to have a smaller MTU?
09:57:55 <priteau> than 1500
09:58:10 <b1airo> priteau: i think they are benchmarking on a tight cluster LAN so MTU *shouldn't* be a major factor (still a factor, but not orders of magnitude)
09:58:15 <oneswig> priteau: correct, it gets trimmed down
09:58:28 <priteau> because of the VXLAN encapsulation inside UDP, if the VM uses 1500 it would cause fragmentation
09:58:34 <b1airo> oneswig: have you seen the MLNX numbers here: https://www.mellanox.com/related-docs/whitepapers/CB_Mellanox_Ethernet_NIC_Advantage.pdf
09:58:58 <b1airo> ah yes good priteau
09:59:01 <b1airo> *good point
09:59:13 <oneswig> b1airo: thanks, I shall got to the webex with those numbers...
10:00:19 <oneswig> OK - I think we must wrap up
10:00:26 <oneswig> Thanks all
10:01:03 <oneswig> I'll gather together the schedule picks and follow up
10:01:13 <oneswig> #endmeeting