21:33:22 <oneswig> #startmeeting scientific-sig
21:33:23 <openstack> Meeting started Tue May  1 21:33:22 2018 UTC and is due to finish in 60 minutes.  The chair is oneswig. Information about MeetBot at http://wiki.debian.org/MeetBot.
21:33:24 <openstack> Useful Commands: #action #agreed #help #info #idea #link #topic #startvote.
21:33:26 <openstack> The meeting name has been set to 'scientific_sig'
21:33:34 <oneswig> sorry y'all
21:33:37 <oneswig> fat fingers
21:33:42 <oneswig> Thanks fungi
21:33:46 <fungi> np
21:34:05 <oneswig> Although how you came by that, I have no idea.
21:34:25 <oneswig> OK, where were we?
21:34:30 <fungi> still had the channel up after the infra meeting
21:34:33 <martial> still on topic 1 :)
21:34:53 <martial> (that said, I must go, will join on my phone)
21:34:54 <fungi> and noticed you seemed to be conducting a meeting with the channel topic still at its (idle) default ;)
21:35:22 <oneswig> omniscience, it's the only rational explanation...
21:35:29 <b1airo> Oh good, so I'm not late :-D
21:35:39 <oneswig> Morning b1airo
21:36:00 <oneswig> technically we only just got started as earlier I stqrted the meeting instead of starting it.
21:36:26 * oneswig is in fine form tonight
21:36:54 <b1airo> Heh
21:38:01 <oneswig> b1airo: only points of note were an etherpad to gather activities (https://etherpad.openstack.org/p/scientific-sig-report-queens) and jmlowe has offered to help book us an evening out for some beer and (perchance) a barbecue in Vancouver.  Similar sort of informal deal to the one in Austin
21:38:28 <b1airo> Oh, awesome. Thanks jmlowe
21:38:45 <oneswig> seconded :-)
21:38:47 <jmlowe> I'll take a crack at it
21:38:54 <oneswig> #topic Scientific OpenStack Constellations
21:39:35 <oneswig> A constellation is a concept being worked on in the TC of a working collection of OpenStack services for meeting a particular use case
21:39:53 <oneswig> eg, containerised workloads or scientific compute
21:40:14 <oneswig> Generalisations are not easy across the broad range of what SIG members do
21:41:00 <oneswig> However, johnthetubaguy has made a bold start for the scientific use case based on our experiences over here.
21:41:28 <oneswig> #link scientific constellation - wip https://review.openstack.org/#/c/565466/
21:41:38 <b1airo> oneswig: reminder we also need to report back to the UC
21:42:11 <b1airo> Yeah I like John's drafting, reviewed it last night after seeing the ML thread
21:42:32 <oneswig> thanks b1airo, sure that will help
21:42:46 <jmlowe> So if I understand this correctly I have a group wanting to host their science gateways on kubernetes, so magnum, barbican, nova, neutron, lbaasv2, octavia, cinder, glance, keystone would be that constellation
21:43:29 <oneswig> jmlowe: I think so, plus some of the glue to make the whole thing coherent.
21:43:47 <b1airo> Kayobe? :-)
21:45:13 <oneswig> Certainly a possibility!  I've got something in mind to set up something like third-party CI for a couple of constellation deployments, as a summer student project, to ensure they stay valid. It's a snapshot otherwise.
21:46:57 <oneswig> It's a star like any other now - https://github.com/openstack/kayobe
21:47:21 <b1airo> Boom tish!
21:47:46 <oneswig> absolutely!  Mark's done a stellar job on it.
21:47:52 <oneswig> (no pun intended0
21:47:54 <b1airo> Sorry, I might be manically tired - pulled a late one
21:48:47 <oneswig> No problem.  We don't have a lot to cover left over.
21:49:02 <oneswig> #topic Ironic RFE specs update
21:49:34 <oneswig> I checked on these two but I don't think there is anything new to report as of this morning.  No matter.
21:50:31 <oneswig> I'm hoping to help the specs through some early iterations, in due course.
21:51:11 <oneswig> #topic AOB
21:52:09 <oneswig> Over here I'm hoping to wrap up the Ceph RDMA investigations I've been doing recently.  One of our systems got an overhaul and it appears this turned off the controlling of cstates.
21:52:40 <oneswig> Even with rados bench working it as hard as possible, the cores are only 10% in C0 and the rest of the time will be revving up and down to C6.
21:52:43 <b1airo> Cstates or Pstates?
21:52:56 <b1airo> Ah right
21:53:02 <oneswig> That's not ideal for a Ceph server
21:53:18 <oneswig> (at least not when it's performance is being measured)
21:53:50 <b1airo> Yeah, see a post from me a couple of years back showing this impact on a buggy tuned throughout-performance profile
21:54:05 <martial_> Metro with a company that does fpga on ssd
21:54:16 <oneswig> So I have some low-hanging fruit to gather on that.  But the RDMA work continues to get interesting, I'm still working on it.
21:54:21 <b1airo> latency-performance was over 30% better
21:54:37 <martial_> Wonder if it could be useful in this case
21:54:37 <oneswig> Thanks b1airo, noted.
21:55:17 <b1airo> Also an interesting thread about Pstates on Skylake just started yesterday
21:55:33 <oneswig> martial_: FPGA on SSD?  Interesting combination.  Recent graphs I've seen of Crosspoint suggest it would go nicely in that mode.
21:55:52 <oneswig> b1airo: on the OpenStack ML?
21:56:36 <b1airo> Ceph users list
21:56:42 <martial_> P
21:56:46 <b1airo> Wido
21:56:57 <martial_> I can try to get them to talk to us if interested
21:57:08 <martial_> I think they were at sc17
21:59:24 <oneswig> #link Ceph-users and frequency scaling http://lists.ceph.com/pipermail/ceph-users-ceph.com/2018-May/026327.html
21:59:42 <oneswig> Looks interesting b1airo, I'll follow it through
22:00:18 <oneswig> martial_: Think that might be a good discussion - but I'm not sure who their customers might be
22:00:26 <oneswig> They can be the judge of that I guess!
22:00:33 <oneswig> Time's up, thanks all
22:00:46 <oneswig> #endmeeting