21:00:16 <oneswig> #startmeeting scientific-sig
21:00:17 <openstack> Meeting started Tue Nov 24 21:00:16 2020 UTC and is due to finish in 60 minutes.  The chair is oneswig. Information about MeetBot at http://wiki.debian.org/MeetBot.
21:00:18 <openstack> Useful Commands: #action #agreed #help #info #idea #link #topic #startvote.
21:00:20 <openstack> The meeting name has been set to 'scientific_sig'
21:00:28 <oneswig> ahoy
21:00:44 <oneswig> happy thanksgiving :-)
21:02:16 <oneswig> The turkey is indeed a powerful force
21:12:31 <trandles> hi oneswig
21:12:38 <trandles> unfortunately super busy today :(
21:13:01 <oneswig> Man, aren't we all.  I'm upgrading Ceph in the other window, how about you?
21:13:31 <trandles> just did an RDO install of Usurri on a testbed...going to do a test OpenShift-on-OpenStack install for a work project
21:14:13 <trandles> sorry I have to run now, just wanted to pop in and say "hi"
21:14:24 <oneswig> Ah, so many questions...
21:14:30 <oneswig> cheerio trandles
21:14:31 <b1airo> morning
21:14:40 <trandles> yeah, me too...maybe I'll have some answers in a couple weeks
21:14:41 <oneswig> Hey b1airo, morning.
21:14:46 <trandles> hi/bye b1airo
21:14:51 <b1airo> irccloud gave me some trouble connecting this morning
21:14:56 <b1airo> hio/bye trandles !
21:15:12 <b1airo> congrats on your confirmed president elect ;-)
21:15:36 <oneswig> b1airo: Seems our usual cohort of Americans are full of turkey :-)
21:16:13 <b1airo> ah yes, that time of year
21:17:01 <oneswig> How's summer coming on?
21:17:33 <b1airo> shit right now :-)
21:17:55 <b1airo> but we've been in the pool already
21:18:12 <oneswig> That's the spirit
21:19:11 <b1airo> got a camper (pop-top) holiday sorted for end of year, so fingers crossed we get a settled couple of weeks
21:19:31 <oneswig> Where will you go?
21:20:27 <b1airo> so what were the big news items out of the Summit and SC? i've been finding it very hard to get much involved in virtual conferences (except where I've been running/presenting something) so feel like most of it has passed me by
21:20:56 <b1airo> heading up top of south island into the golden bay area and then over onto the top of the west coast there
21:21:00 <oneswig> I got the free-tier ticket for SC but didn't actually use it.
21:21:26 <oneswig> Nelson?  Abel Tasman?  I remember it well(ish)
21:22:32 <oneswig> I've had a fair bit going on over here so haven't been back through the Open Infra Summit talks.
21:22:58 <oneswig> The big thing for us at SC was getting the Ironic system in at #99 in the TOP500
21:23:01 <b1airo> yeah, i need to queue up my youtube backlog too
21:23:16 <oneswig> 0.001 PF more than #100...
21:23:49 <b1airo> Julia must have been pretty happy to hear about that
21:24:02 <oneswig> I hope so.
21:24:37 <oneswig> You travelling much internally still?
21:26:02 <b1airo> no quite static at the moment. probably one, maybe two trips, before the year is out - but think i'm only at 4 or 5 for the year. one of those was only a day in and auckland went back into lock-down so i high-tailed it out of there
21:26:36 <b1airo> will depend how Bel's concussion going too :-/
21:26:41 <oneswig> It's all over Bristol right now.
21:26:47 <oneswig> How's she doing?
21:28:37 <b1airo> feel like you're taking a gamble every time you go to the shops? must be stressful
21:29:45 <b1airo> she's pretty good considering she probably hit her head on a "sleeper" (ha!) from 2-3m drop and had temporary dysphasia
21:29:50 <oneswig> Not too much.  The kids both go to school and my wife works in the local uni.  Infection rates locally are high (~500 per 100k people) but even so it's not that bad.
21:30:26 <oneswig> 2-3m drop?  Ouch.  Bet that gave her a good bump.
21:30:28 <b1airo> is that confirmed infections or based on modelling accounting for asymptomatic cases?
21:31:27 <oneswig> I think it's positive tests, so underestimating true prevalence rates.
21:32:31 <oneswig> Everyone's locked down right now so things are settling again.
21:34:07 <b1airo> left field question... have you ever used NVIDIA MPS ?
21:34:22 <oneswig> MPS?  Not heard of it.
21:35:28 <oneswig> I'm not a user of GPUs, particularly
21:35:31 <oneswig> More a plumber.
21:36:02 <b1airo> ha
21:37:06 <b1airo> we've got a few (8) A100s on the way and i'm thinking about how we can maximise utilisation across them
21:37:32 <oneswig> They aren't all in one box are they?
21:38:01 <b1airo> GPU underutilisation seems to be incredibly common but no one ever talks about it...
21:38:33 <b1airo> no, going to be 2 per node (with NVbridge across them)
21:40:20 <b1airo> we'll go denser once/if we see sustained demand for that
21:40:51 <oneswig> It's a challenge I'm sure to make the most of an exotic resource.
21:42:15 <oneswig> b1airo: ever converted a Ceph deployment from non-containerised to containerised?  We've used the ceph-ansible playbook for this previously.  I'll be watching it very closely though.
21:42:43 <b1airo> well that's the thing though, they're not that exotic anymore, insofar as there are plenty of GPU accelerated codes out there. but the obvious issue is that not many of those codes use GPU at a sustained rate
21:43:27 <oneswig> right, I see.
21:44:05 <b1airo> no, though it doesn't sound fun. presumably the main challenge is making sure you don't have to redeploy the OSDs and replicate data across a 3 month upgrade window
21:44:06 <oneswig> You want applications that yield the GPU when they aren't using it.  That may be a long shot
21:44:36 <b1airo> that's one of the things that MPS does - cooperative GPU sharing for CUDA workloads
21:44:39 <oneswig> Thankfully the cluster is smallish and fast
21:45:21 <b1airo> ah right, so you are blowing away the extra OSD data?
21:46:12 <oneswig> Upgrade Ceph, containerise, and redeploy with CentOS 8.  In that order.
21:47:38 <oneswig> It's quite a test plan
21:49:36 <b1airo> hmm yeah, i guess the theory is that everything will be much easier once it is containerised...
21:50:44 <oneswig> I am not certain about how much easier it will be.  The hardest bit will be the CentOS reinstall.
21:51:00 <b1airo> the thing we always forget when getting excited about software defined infrastructure - software sucks! :-)
21:51:23 <oneswig> always a bit too clever for its own good
21:52:33 <b1airo> ok, i'm going to run and see if i can make sure we're sorted this end for our Friday morning session
21:52:59 <oneswig> sounds good.  Until then b1airo
21:53:06 <b1airo> cheers
21:53:13 <oneswig> cheers!
21:53:16 <oneswig> #endmeeting