21:00:16 <oneswig> #startmeeting scientific-sig
21:00:17 <openstack> Meeting started Tue Sep  1 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> Hey Scoob
21:01:03 <oneswig> #link Today's agenda https://wiki.openstack.org/wiki/Scientific_SIG#IRC_Meeting_September_1st_2020
21:01:14 <TheJulia> o/
21:01:31 <oneswig> Hi TheJulia, welcome!
21:01:42 <oneswig> ears burning? :-)
21:02:27 <oneswig> martial: you there?
21:02:29 <trandles> Hi all
21:02:37 <oneswig> Hi trandles
21:02:38 <martial> yes Hi :)
21:02:47 <oneswig> #chair martial
21:02:48 <openstack> Current chairs: martial oneswig
21:02:53 <TheJulia> oneswig: no, not really burning :)
21:03:21 <oneswig> awesome, well let's talk kexec...
21:03:27 <oneswig> #topic Ironic deploy and kexec
21:04:04 <oneswig> context on this is that BIOS is sometimes just too damn slow to endure a second lap on deployment
21:04:56 <oneswig> In some circumstances, we are measured by the time taken to deploy / redeploy n servers where n is uncomfortably large.
21:05:13 <TheJulia> I've wanted to see this be a thing in ironic for quite some time. However I simply don't have time to make it happen on my own. I've got some patches, and I'm happy to hack on patches occasionally, but could really use some help with at least some reviews and testing.
21:05:46 <TheJulia> So in our midcycle call this week, the topic came up of trying to get a smallish working group together of interested parties to collaborate and try and push it over the finish line.
21:06:01 <TheJulia> The added bonus of which, being able to influence use cases/operation and such
21:06:28 <oneswig> TheJulia: I hope someone from our team can help.  Worst case, its me...
21:06:40 <trandles> I'm on board for testing/review/QA
21:06:42 <TheJulia> That would be awesome!
21:07:03 <TheJulia> Anyway, I created a quick dooddle with some time windows to try and see if we could get on a call to discuss.
21:07:06 <TheJulia> #link https://doodle.com/poll/ihbbtns36b2rkhfd
21:08:12 <TheJulia> I guess that is kind of it unless there are questions/comments/thoughts/wants/etc
21:08:22 * TheJulia adds needs to that list
21:08:43 <oneswig> Great.  Where are you at with your patches?
21:10:17 <TheJulia> oneswig: need rebasing, they haven't been tested, more so just trying to get the idea out of my head
21:10:33 <oneswig> One other thought - is this compatible with multi-tenant networking, or are they mutually exclusive due to the shortcuts taken?
21:10:41 <TheJulia> mostly to see the areas where things would tie in
21:11:39 <TheJulia> multi-tenant networking would need tobe something we consider since it is another path. It might be feasible but we may jsut need to be able to tell ironic's agent to "go kexec in 15 seconds"
21:12:07 <TheJulia> naturally there is also greater risk there, but we're at the stage where discussion of such needs/wants/concerns/thoughts/desires are important
21:13:20 <oneswig> It does seem tricky...
21:13:56 <TheJulia> Yeah, I also kind of want to see kexecing from a persistent agent as well, so it does lend itself to needing to being able to navigate multiple paths
21:14:30 <oneswig> Sorry What do you mean by persistent agent?
21:14:37 <TheJulia> And yes, I'm thinking some operators could go from a deployed workload, to deployment ramdisk, to another workload without an actual reboot in such a case
21:15:26 <trandles> I'm hoping standalone is the easiest case and where the work starts (mostly because that's the one I need)
21:15:28 <oneswig> OK - wow, interesting idea.
21:15:28 <TheJulia> there has been some community interest in trying to eventually make ironic's agent such that it could live inside an instance and eventually tear down the node from with-in. Naturally there are other concerns there that would need to be navigated, but it is a neat concept of sorts
21:15:50 <TheJulia> trandles: ++
21:18:06 <oneswig> TheJulia: what's the feature deadline for Victoria?  Too soon?
21:18:49 <TheJulia> It is very unlikely to make Victoria. If there is enough interest it should make our first release after the end of victoria cycle release
21:19:13 <TheJulia> of course, if people are really interested and want to focus on it, maybe we might be able to make things happen in a couple weeks
21:19:30 <oneswig> That would be a stretch at this end.
21:19:36 <TheJulia> Unfortunately I lack a time machine to go collect the code from the future as well...
21:20:21 * TheJulia hopes someone got a chuckle out of that
21:20:35 <oneswig> I would lend you mine but Trump borrowed it and hasn't given it back...
21:20:37 <trandles> I'll expedite figuring out the upstream stuff so I can at least review in a meaningful way
21:20:51 <TheJulia> oneswig: DOH!
21:21:21 <TheJulia> trandles: I'd be happy to jump on a call and kind of try to walk you through that if you want
21:21:22 <trandles> if Trump is working on future knowledge we're all doomed...nevermind TheJulia we won't need kexec
21:21:51 <oneswig> The tangled webs we weave.
21:22:01 <TheJulia> silly paradoxes :(
21:22:16 <trandles> TheJulia: thanks for the offer. I'll ask for help if I get stuck.
21:22:41 <oneswig> TheJulia: will you announce the kexec-wg meeting on openstack-discuss? I'll keep an eye out if so.
21:24:07 <TheJulia> oneswig: I will, and I'll send a calendar invite out to participants if I have their email addresses
21:24:09 <TheJulia> or can find them
21:24:48 <oneswig> sounds good to me.  Thanks TheJulia
21:26:29 <oneswig> OK - any more on kexec?
21:27:10 <TheJulia> Not from me, thanks!
21:27:23 <oneswig> Great, thanks for coming along TheJulia
21:27:35 <oneswig> #topic CANOPIE-HPC
21:27:48 <oneswig> Busy week for the LANL folks it seems
21:27:59 <trandles> always I'm afraid
21:28:02 <oneswig> trandles: are you involved in CANOPIE?
21:28:13 <trandles> I'm a reviewer so beware
21:28:39 <trandles> well, I said "yes" when asked but haven't heard anything more
21:28:41 <oneswig> #link CFP for CANOPIE https://www.canopie-hpc.org/cfp20/
21:29:07 <oneswig> What's the focus this time around, anything specific?
21:29:19 <trandles> I think the new deadline is Sept. 10 btw
21:29:49 <trandles> oneswig: checking...
21:31:00 <trandles> no particular theme that I know of, quite a broad scope at https://www.canopie-hpc.org/cfp20/
21:31:36 <oneswig> As a counterpoint I'm trying to think of a topic that's out of scope :-)
21:32:03 <trandles> the CFP does say "include but are not limited to"
21:32:15 <oneswig> Damn, headed that one off!
21:33:12 <oneswig> Debugging's an interesting subject.  Can you attach gdb to a process in a container and expect gdb to find the exec and symbols?
21:33:40 <trandles> that is a topic I'd love to hear more about, not just debuggers but analysis tools like tau, stat, etc.
21:34:18 <trandles> it's on my own internal hot list of things the community needs to get a handle on and develop some best practices
21:34:31 <martial> Deadline has been extended to 9/10
21:35:29 <trandles> also best practices for containerized MPI applications at scale
21:35:45 <trandles> Charliecloud has spoiled me, other runtimes are painfully horrible at that
21:36:23 <oneswig> What's new with Charliecloud?
21:36:46 <trandles> unpriviledged build
21:36:56 <trandles> totally unprivileged, no setuid nonsense
21:37:29 <trandles> also on the near term roadmap, multi-stage builds
21:37:39 <trandles> better support for `spack containerize`
21:40:38 <trandles> there are lots of other things but maybe not of general interest...like all containers are read-only and if you application can't handle that you're doing it wrong
21:40:45 <trandles> *your
21:41:02 <oneswig> On a tangent we've had huge problems building and running WRF with any semblance of portability.  I'd buy so much beer for someone who could containerise it in a way that works.  Unfortunately it's a mutant in that the data model somehow gets compiled into the code.  You think you've got badly-behaved applicatoins :-)
21:41:06 <trandles> *your appliction*...geesh
21:41:30 <trandles> I'm not familiar with WRF
21:41:53 <trandles> oh, weather, nevermind
21:42:00 <oneswig> weather code.  When we have a client who wants it, it's always a stormy forecast
21:42:18 <trandles> I'll buy you a beer for that one ;)
21:42:26 <martial> lol
21:43:24 <oneswig> trandles: do you use the tau visualization tool?
21:43:29 <trandles> we do
21:43:56 <trandles> beyond that, it's far from my area
21:45:02 <oneswig> must be a fascinating field to work in.
21:45:35 <trandles> we just had a tau workshop for users last week
21:47:52 <oneswig> their example of visualising MPI_Alltoallv over 786432 cores ...
21:48:07 <oneswig> Such a shame they couldn't round up to the million
21:48:48 <martial> have you seen the new gaming GPU announced today?
21:49:38 <oneswig> No, what's new?
21:49:50 <trandles> RTX 3090?
21:50:01 <martial> they announced the 3070, 3080 and yes the 3090
21:50:17 <martial> we all kind of knew it was coming :)
21:50:27 <martial> but it is a nice range of hardware
21:51:21 <martial> they announced a few other things, but the 3090 will have 24GB of memory
21:51:48 <martial> if you want to see the announcement, here is the youtube link https://www.youtube.com/watch?v=QKx-eMAVK70
21:52:20 <martial> no Ti this time
21:53:44 <oneswig> Sounds ideal for business graphics.
21:55:04 <martial> yes it looks rather nice
21:55:22 <martial> Performance is 1.5 to 2 times faster than previous gen
21:55:26 <martial> which is nice as well
21:56:01 <trandles> I could say "what's a GPU?" because other than my work laptop I don't have a computer at home. :P
21:56:32 <martial> I should not start to count the number of computers in this room alone :)
21:56:45 <trandles> I use my phone for everything...which says a lot about the current state of HPC hardware :(
21:57:16 <martial> hey it works for what you need
21:57:28 <trandles> when nvidia buys ARM I'm going to look into llama farming
21:57:29 <oneswig> trandles: those huawei phones are awesome :-)
21:58:23 <martial> or one plus
21:58:31 <oneswig> Nearly at time - last gasp comments
21:58:45 <trandles> PTG?!?!
21:58:50 <martial> trandles that is a plan
21:58:53 <martial> oh right :)
21:58:57 <trandles> is the SIG submitting something?
21:59:02 <martial> #topic PTG
21:59:17 <martial> we were planning to
21:59:19 <oneswig> ah yes.  I took an action two weeks ago to do something like last time.
21:59:25 <martial> I know we discussed it
21:59:25 <trandles> martial will do it, great, thanks Martial!
21:59:43 <martial> did I just walk into a trap?
22:00:03 <martial> I am pretty sure Stig is mostly done :)
22:00:09 <oneswig> A trap so subtle
22:00:28 <martial> Stig: did you do something or not?
22:00:31 <oneswig> np, I said I'd do the paperwork - I will sort it.
22:00:40 <martial> I can take care of it, but not before next week
22:00:49 <martial> what is the deadline?
22:00:59 <oneswig> 9th? 11th?
22:01:08 <oneswig> I'll look into it tomorrow.
22:01:16 <oneswig> time to close
22:01:21 <martial> thank you; ping me if you need help
22:01:23 <trandles> I'd do it but I'm busy working on containerizing WRF for oneswig. He offered beer.
22:01:37 <oneswig> trandles: you don't know what you're getting into, walk away
22:01:38 <martial> hey ... the right priorities :)
22:01:53 <oneswig> OK, thanks all
22:01:55 <oneswig> #endmeeting