21:00:30 <martial> #startmeeting scientific-sig
21:00:31 <oneswig> ah - no '-'
21:00:32 <openstack> Meeting started Tue May 11 21:00:30 2021 UTC and is due to finish in 60 minutes.  The chair is martial. Information about MeetBot at http://wiki.debian.org/MeetBot.
21:00:33 <openstack> Useful Commands: #action #agreed #help #info #idea #link #topic #startvote.
21:00:35 <openstack> The meeting name has been set to 'scientific_sig'
21:00:42 <oneswig> oof1
21:01:07 <martial> #chair oneswig
21:01:08 <openstack> Current chairs: martial oneswig
21:01:29 <martial> Welcome friends to yet another instance of the Scienfitic SIG meeting :)
21:01:29 <oneswig> Hey
21:01:31 <trandles> hi oneswig martial
21:01:38 <martial> hello Tim
21:01:39 <oneswig> Hi trandles
21:02:07 <oneswig> I've had a couple of quality hours trying to comprehend a legal agreement...
21:02:25 <martial> ah?
21:03:27 <trandles> that doesn't sound like fun oneswig
21:03:31 <oneswig> One of the fun aspects of working in a small firm is not having a dedicated legal department.  Living the dream :-)
21:04:16 <oneswig> It's actually not too bad - which is worrying me.
21:04:30 <martial> was looking for our agenda for today and I guess freeform meeting it is :)
21:04:48 <b1airo> morning
21:05:20 <oneswig> Hi b1airo, morning
21:05:34 <b1airo> how goes it? (legal reviews aside)
21:05:36 <martial> welcome Blair
21:05:46 <martial> #chair b1airo
21:05:47 <openstack> Current chairs: b1airo martial oneswig
21:07:54 <oneswig> Anyone come across this in the line of work? https://ronin.cloud
21:08:08 <b1airo> what was the wash up from the PTG? anything I should be staying on top of from a system architect perspective?
21:08:08 <oneswig> Asking for a friend...
21:09:18 <oneswig> From the Nova perspective I think it was good to see continued work on mixing "hpc" and "general purpose" workloads
21:09:46 <oneswig> mostly better ways to share pinned and non-pinned workloads
21:12:07 <oneswig> This work in Wallaby will come in handy for smaller deploys and combining GPU and non-GPU workloads in a way that doesn't bisect the cloud: https://docs.openstack.org/nova/wallaby/admin/managing-resource-providers.html
21:13:19 <b1airo> i think i have seen Ronin before, but it wasn't as polished IIRC
21:16:50 <b1airo> anyone other than Mike integrated vGPU yet, assuming oneswig and martial ?
21:17:10 <martial> no vGPU for me yet
21:18:09 <oneswig> b1airo: when given the option, clients tend to stop short of the effort (and cost).  We have a couple of new deploys with GPUs that may evaluate it but no concrete timing on either.
21:18:35 <oneswig> One is V100 and the other is A100, which may make it interesting when comparing overhead.
21:19:04 <trandles> I'm in the middle of doing an RHOSP deploy to support an OpenShift project, including bare metal nodes to support containerized windows applications.
21:19:16 <trandles> ^ this is another way of saying "I've been drinking heavily"
21:19:17 <oneswig> The A100 is more likely to go MIG I think
21:19:43 <oneswig> trandles: containerized windows applications on OpenShift?  Sheesh!
21:20:08 <oneswig> I think it's time we saw Charliecloud on OpenShift.
21:20:58 <trandles> It's challenging enough that we have daily 4.5 hour work sessions with RHAT engineers for the next 8 weeks.
21:21:42 <oneswig> Daily!  What about all the other stuff you were doing in your day? :-)
21:22:10 <oneswig> Although I'm actually quite envious at the idea of having 4.5 hours to focus on one thing every day
21:23:00 <trandles> Luckily I'm just supporting someone else in the daily sessions. It's not my 4.5 hours.
21:23:37 <b1airo> trandles: :-D
21:24:28 <oneswig> How did you arrive at 4.5 hours?  Just curious
21:25:34 <trandles> probably the amount of money...??  Not sure. I'm trying to stay as disengaged as possible while still being a meaningful participant :P
21:25:42 <martial> what's the expected outcome?
21:26:36 <trandles> martial: the driver of the project is way above me. By the time it got on my radar it was a task list.
21:26:49 <trandles> 1. get RHOSP install
21:26:57 <trandles> 2. get OpenShift working
21:27:10 <trandles> here's a bunch of people to work with, you've probably never met them
21:28:37 <oneswig> A familiar tale.
21:29:41 <oneswig> It goes comprehensively off piste when you bring in the windows containers.
21:30:55 <martial> wait why windows?
21:30:56 <b1airo> yeah, i foresee a great "Curious tale of ..." story coming out of this
21:31:49 <trandles> The project needs to support a whole boatload of applications. Whoever made the decision didn't bother to survey the applications they said _must_ be moved to containers.
21:32:07 <b1airo> re. vGPU oneswig - you mentioned cost, are you talking about the NVIDIA licensing aspect? it's basically free for academic usage iiuc
21:32:20 <martial> ouch ... another "digital modernization" project
21:33:53 <b1airo> and re MIG, well yes but that's kind of orthogonal - if you fundamentally want (v)machine level isolation with GPU sharing, vGPU is the only option
21:34:48 <b1airo> though admittedly some of the killer use-cases could be implemented with MIG + K8s
21:34:51 <oneswig> b1airo: yeah the VCS license.
21:35:36 <b1airo> i'm trying to piece together whether it's going to work on a CentOS/Stream based platform...
21:35:55 <martial> anybody tested RockyOS btw?
21:36:33 <oneswig> Sounds like you've had better luck getting good answers from Nvidia on the implementation details.  VGPU has been described to me as including time-sliced sharing which sounded concerning.
21:36:41 <b1airo> seems like probably, but would be nice to hear from someone who has done it. i'm assuming JetStream is CentOS based - I'll followup with Mike in Slack...
21:36:49 <oneswig> martial: not looked at it.  We've been testing stream out.
21:37:16 <b1airo> is there actually a Rocky release now? what about Alma...?
21:38:59 <b1airo> yeah but the nice thing about vGPU compared to MIG (if I'm remembering this correctly) is that any one individual VM can get all the compute power of the GPU if you want to allow them to burst (while other VMs are not using much/any GPU)
21:39:17 <martial> b1airo: RC1
21:39:47 <b1airo> any inside gossip on the Singularity fork... trandles ?
21:40:14 <b1airo> ok, maybe not "inside gossip", more insight I guess
21:40:33 <oneswig> b1airo: what have I missed - fork?
21:41:53 <b1airo> https://sylabs.io/2021/05/singularity-community-edition
21:44:11 <oneswig> "When two entities are equally as passionate about something but see things differently, it has the potential to cause unnecessary tension, toxicity or collateral damage" - I don't think the project would have forked simply at the potential for that, is my guess.
21:50:26 <b1airo> yeah, much as i cringe at the earnest language in the rest of the article i feel like that was probably a generous take
21:52:08 <trandles> b1airo: I have no inside info on Singularity
21:52:34 <oneswig> trandles: what's the latest on CharliecloudCE? :-)
21:54:06 <trandles> although the letter from Kurtzer to Sylabs saying "you forked Singularity, rename it please" made me chuckle
21:54:49 <oneswig> Is Duality taken I wonder?
21:55:40 <trandles> He started Singularity, he started Sylabs, he sold Sylabs, he started HPCng, HPCng pissed off Sylabs enough that Sylabs forked it, Kurtzer asked Sylabs to rename their fork, he regains control of "Singularity?"
21:59:45 <oneswig> Time's nearly up!
21:59:49 <martial> reaching the top of the hour
22:00:02 <martial> thanks everybody for joining
22:00:07 <martial> #endmeeting