11:00:12 <oneswig> #startmeeting scientific-sig
11:00:13 <openstack> Meeting started Wed Feb 13 11:00:12 2019 UTC and is due to finish in 60 minutes.  The chair is oneswig. Information about MeetBot at http://wiki.debian.org/MeetBot.
11:00:14 <openstack> Useful Commands: #action #agreed #help #info #idea #link #topic #startvote.
11:00:17 <openstack> The meeting name has been set to 'scientific_sig'
11:00:24 <b1airo> evening
11:00:26 <oneswig> Greetings Earthlings
11:00:30 <oneswig> #chair b1airo
11:00:31 <openstack> Current chairs: b1airo oneswig
11:00:46 <janders> good evening
11:00:47 <oneswig> #link Agenda for today https://wiki.openstack.org/wiki/Scientific_SIG#IRC_Meeting_February_13th_2019
11:00:55 <b1airo> hmm, never noticed the info command before, what does that do...?
11:00:58 <oneswig> Hi janders
11:01:00 <b1airo> #info
11:01:06 <b1airo> #help
11:01:22 <b1airo> #help info
11:01:22 * oneswig wonders what that big red button does
11:01:28 <janders> boom!
11:01:32 <janders> great balls of fire
11:01:34 <b1airo> push it! push it!
11:01:53 <oneswig> ha.  Kids must have coloured in the light switch again :-)
11:02:07 <b1airo> yes, i subscribe to the "don't ask for permission..." doctrine
11:02:35 <b1airo> i think there's something about maybe apologising afterwards
11:02:57 <oneswig> A good mantra, except possibly when driving.
11:03:07 <janders> or handling nukes
11:03:17 <oneswig> so true!
11:03:40 <b1airo> my nana had a "vertical takeoff" button in one of her cars i remember quite clearly. had disappointing results, i.e., nothing
11:04:15 <oneswig> Over here we had a research computing cloud conference yesterday, was a good session as usual
11:04:20 <janders> a friend had a star trek voice in his GPS. Instead of "arriving at destination" it had "initiate docking sequence"
11:04:42 <b1airo> heh, nerd status plus
11:04:44 <oneswig> janders: So impressive, except possibly when going on a first date.
11:04:56 <verdurin> Morning.
11:05:05 <oneswig> Hi verdurin, morning
11:05:17 <oneswig> Good work yesterday
11:05:30 <verdurin> Yes, I thought it was a good day.
11:05:52 <oneswig> Me too.  We should perhaps compare notes in AOB.
11:06:06 <janders> regarding research computing & cloud - does AlaSKA ingest data straight from Telescope/FPGAs?
11:06:10 <b1airo> sounds intriguing
11:06:20 <janders> (was chatting to Radioastronomy guys this arvo)
11:06:25 <oneswig> janders: only simulated ones.
11:06:43 <janders> I was wondering if radioastronomy is a good use case for dual-interconnect bare-metal clouds
11:07:01 <janders> (ingest via 100GE/ROCE, write to IB parallel FS)
11:07:04 <oneswig> We generate artificial streams of UDP data ("SPEAD" packets) due to our lack of actual telescope
11:07:21 <oneswig> janders: exactly so.
11:07:40 <b1airo> it's all just white noise anyway, amirite?
11:07:54 * b1airo ducks
11:08:03 <oneswig> b1airo: the noise is far, far louder than the signal
11:08:12 <oneswig> There was that famous snowflake analogy
11:08:23 <janders> I will see if we can find some common ground.. I think they might really need some kit in the middle of WA desert - which might be doable but not sure how popular with the execs
11:08:49 <oneswig> https://www.goodreads.com/quotes/865839-the-total-amount-of-energy-from-outside-the-solar-system
11:08:54 <janders> edge computing is a hot topic and 1Tbps links are rather expensive
11:09:33 <janders> no snowflakes in Western Australia, I don't think
11:10:06 <oneswig> janders: kit in the desert is a desperate measure, not to be taken lightly, from what I've heard.
11:10:12 <ildikov> edge computing is indeed a hot topic, I'm happy to learn if you have a use case :)
11:10:32 <oneswig> Hi ildikov!
11:10:38 <ildikov> oneswig: hi :)
11:10:40 <oneswig> Many, I'd say
11:10:47 <ildikov> I'm just lurking
11:10:56 <oneswig> Lurk away :-)
11:11:16 <oneswig> Next topic might benefit from some Foundation input
11:11:17 <ildikov> also co-leading the edge working group and we're always hungry for cool new use cases!
11:11:43 <oneswig> #topic OpenStack at ISC
11:11:58 <janders_> ildikov: good to know
11:12:04 <b1airo> nothing cool about a containerised data-centre in the middle of Western Australia desert, except the concept and science, obviously
11:12:08 <oneswig> ildikov: perhaps we should set aside a session to discuss edge use cases
11:12:20 <janders_> +1
11:12:29 <ildikov> sounds good to me
11:12:44 <b1airo> yeah, seems very relevant for research/science orgs
11:12:51 <ildikov> we have some examples here: https://wiki.openstack.org/wiki/Edge_Computing_Group/Use_Cases
11:13:11 <oneswig> OK, there's a BoF opportunity to gather people attending ISC in Frankfurt in June
11:13:20 <oneswig> Deadline next Wednesday (20th February)
11:14:00 <oneswig> It would be good to gather panelists for a BoF proposal
11:14:52 <janders_> I don't think it's likely I will make it to the ISC but I can ask around the other teams who's going
11:14:56 <oneswig> We've contacted a couple of people from EOSC and ELIXIR federations, but I think we need 5 or so people to put their names to it.
11:15:21 <ildikov> I support the idea
11:15:32 <oneswig> janders_: please do and if I get a response, mail it due to the tight timing.
11:15:56 <janders_> I just texted the main suspect
11:16:03 <oneswig> ildikov: great - would you or anyone else from the Foundation be interested in attending?
11:16:18 <oneswig> Frankfurt's practically your back yard :-)
11:16:32 <ildikov> oneswig: do you have a link with info about ISC?
11:16:52 <oneswig> #link https://www.isc-hpc.com/bof-sessions-2019.html
11:16:54 <ildikov> I need to check in with the team before committing, but it's indeed an easy trip
11:16:56 <b1airo> looks to me like a good opportunity to engage potential new community members, and so far as i can tell OpenStack hasn't figured on the programme yet
11:16:56 <oneswig> That's the one
11:17:19 <ildikov> oneswig: I'll get back to you on it this week. Would that work?
11:17:32 <oneswig> b1airo: agreed.  It's a good way to find the people on the HPC fringes of our space.
11:17:38 <oneswig> ildikov: yes, that would be great.
11:17:55 <oneswig> b1airo: Are you going on behalf of NESI?
11:18:15 <ildikov> oneswig: cool, will let you know as soon as I can
11:18:16 <b1airo> i will if we run this
11:18:55 <oneswig> fantastic.
11:19:18 <oneswig> I'll ask DK if he's going to be there, he usually has plenty of interesting things to talk about.
11:19:19 <b1airo> until someone blocks my expense account anyway...
11:19:35 <b1airo> DK Panda?
11:19:39 <oneswig> b1airo: I assume that's what happened last time? :-P
11:19:43 <oneswig> b1airo: yes
11:20:21 <b1airo> he ran a BoF last year, so seems likely he'll be proposing same again: https://2018.isc-program.com/presentation/?id=bof126&sess=sess186
11:21:19 <b1airo> ha, just noticed Christian Kniep was on that one too. bit of a small world this isn't it..
11:21:42 <oneswig> That's the one, I think you're right about the small world.
11:22:03 <oneswig> We should try to broaden it :-)
11:22:16 <b1airo> +1
11:22:55 <oneswig> If anyone else here is interested in participating, or could suggest someone to invite, please let me know
11:23:31 <oneswig> I'll put together some wording based on the SC BoF proposal.
11:23:38 <oneswig> Any more to add?
11:23:56 <b1airo> was just about to say the same thing, actually just digging for the last SC one now...
11:25:15 <oneswig> Thanks b1airo
11:25:32 <oneswig> #topic Update on Summit/Forum/PTG
11:26:21 <oneswig> Martial has booked us a half-day session at the PTG.  Not sure which day but towards the end of the week.
11:26:39 <oneswig> I'm hoping for Thursday (I need to head home Thursday night...)
11:27:28 <oneswig> We'll apply for the usual format of meeting + BoF/lightning talks as well.
11:28:52 <oneswig> I don't think there was anything to add on that update for now.
11:29:02 <janders_> sounds good!
11:29:09 <oneswig> #topic Manila + GPFS user experiences
11:29:48 <oneswig> So I found somebody who has tried the Manila driver for GPFS
11:29:55 <oneswig> And had a few minutes discussion with them about it.
11:30:10 * b1airo grabs popcorn
11:30:18 <oneswig> #link This report makes a fairly neutral reference to it https://cloud.ac.uk/reports/spectrumscale/
11:30:53 <oneswig> The people who wrote it are a team from Birmingham University in the UK (Simon and Radoslaw)
11:31:23 <oneswig> They did an evaluation on Manila + GPFS about 18 months ago
11:31:56 <janders_> great find, thank you
11:32:00 <oneswig> Their conclusion was that the plugin doesn't add much value.  It had some limitations, in particular requiring flat networking.
11:32:01 <b1airo> ah yes, i now remember speaking to Simon (presumably at a Summit)
11:32:26 <verdurin> Yes, as I recall the verdict was not all that positive.
11:33:58 <oneswig> The model of re-exporting NFS via Ganesha is not the most efficient.  Simon referred to 3 tiers of memory buffering to tune and balance - Ganesha, GPFS, kernel buffer cache.  He thought there might be more promise in automating the management of CES (Cluster Export Services) within GPFS to achieve a similar effect.
11:34:11 <janders_> I haven't used Manila in my GPFS days, so this sounds possible (and insane)
11:34:20 <janders_> flat networking requirement..
11:34:22 <oneswig> Although he was wondering if multiple instances of CES (one per tenancy) would work from a single GPFS instance
11:34:57 <janders_> sigh sounds like there isn't many options for RDMA storage for OpenStack :(
11:35:45 <oneswig> Radoslaw pointed out the documentation has not updated since 2017 (think we spotted that last time).  It does seem like this driver is in maintenance mode at best.  Simon mentioned he thought the team at IBM were working on other projects and not giving it attention.
11:36:37 <oneswig> I guess the choice for it would be to take on refreshing the driver and adding the features desired - or do something outside of OpenStack's purview.
11:36:39 <janders_> ..and to add to this, I heard goss that IBM are pulling out from OpenStack/GPFS work to focus on k8s integration
11:36:44 <verdurin> Yes, agree with that impression from IBM.
11:37:12 <janders_> given OpenStack and K8S are quite tightly coupled at times, it's an interesting statement
11:37:14 <b1airo> would CES be the only path to multi-tenancy, i.e., secure sub-tree isolated exports/mounts ?
11:37:52 <oneswig> b1airo: I don't know enough about GPFS to answer that but why don't I put you in touch with the horses mouth (so to speak).
11:38:01 <verdurin> b1airo: there's a reason we're now looking at Lustre to complement our GPFS.
11:39:04 <oneswig> I didn't think to ask if they knew of anyone else who has tried this driver.
11:39:40 <b1airo> fair enough verdurin . having to manage it via lnet seems a bit icky to me though
11:40:01 <b1airo> thanks oneswig
11:40:05 <verdurin> b1airo: there's no ick-free approach, sadly.
11:40:15 <janders_> +1 :(
11:40:20 <oneswig> no problem, a shame not to find a live user though
11:40:54 <oneswig> OK, thats all I had on it.
11:41:04 <oneswig> #topic AOB
11:41:36 <oneswig> That Docker vulnerability.
11:42:46 <oneswig> #link https://www.theregister.co.uk/2019/02/11/docker_container_flaw/
11:43:33 <oneswig> It's not quite as straightforward as writing to /proc/self/exec - there's some kind of race condition loophole that must be exploited, it seems
11:45:06 <oneswig> One of our team was doing a demo of Magnum on bare metal yesterday.  It was exciting enough as it was but I haven't seen if there are patched Magnum root images available yet.
11:46:22 <oneswig> verdurin: what was your impression of the day yesterday?
11:46:55 <b1airo> on that note, can you configure Magnum to dedicate a host to only one pod?
11:47:35 <oneswig> b1airo: will ask
11:47:37 <verdurin> oneswig: as Phil said, it's good to see the talks evolving to what you might call a higher-level view
11:48:08 <verdurin> Was impressed with the demos in the session I chaired.
11:49:27 <oneswig> me too, there were some useful things emerging.  And entirely new ways of deploying Slurm in the cloud as well :-)
11:50:55 <oneswig> There did seem to be a lot more attention from public clouds to meeting HPC requirements.
11:51:36 <verdurin> Yes, more depth in their approach than before.
11:51:37 <b1airo> which were those?
11:51:46 <janders_> new ways of deploying Slurm... what's the best example or two?
11:52:01 <janders_> always interesting to hear
11:52:46 <oneswig> #link this was one - Terraform for Oracle cloud https://cluster-in-the-cloud.readthedocs.io/en/latest/
11:53:44 <janders_> terraform - nice! :)
11:54:20 <oneswig> The interesting piece of this work was scheduling different job steps to different resource classes (eg, GPU, high memory).  One possible flaw was that nodes were provisioned specifically to run each job.
11:55:02 <oneswig> janders_: I wanted to ask, why the project is limited to one cloud, and whether it's possible to write portable Terraform.
11:55:36 <oneswig> Be good to hear a data point on that from someone with extensive experience of trying
11:55:54 <janders_> indeed! :)
11:56:00 <verdurin> oneswig: yes, my impression is that running Terraform across multiple clouds isn't as straightoforward as you would hope
11:58:01 <oneswig> Perhaps I'll find the presenter and ask him.
11:58:17 <b1airo> the job step mapping thing sounds like documentation we need on our systems
11:58:43 <b1airo> other commercials there?
11:59:08 <oneswig> Google and Azure, don't think I saw anyone from AWS
11:59:38 <oneswig> OK we are nearly out of time, any more?
12:00:09 <oneswig> time at the bar - thanks everyone
12:00:12 <oneswig> #endmeeting