09:00:45 <oneswig> #startmeeting scientific-wg
09:00:46 <openstack> Meeting started Wed Mar  1 09:00:45 2017 UTC and is due to finish in 60 minutes.  The chair is oneswig. Information about MeetBot at http://wiki.debian.org/MeetBot.
09:00:47 <openstack> Useful Commands: #action #agreed #help #info #idea #link #topic #startvote.
09:00:49 <openstack> The meeting name has been set to 'scientific_wg'
09:01:03 <oneswig> Greetings
09:01:42 <oneswig> #link agenda for today https://wiki.openstack.org/wiki/Scientific_working_group#IRC_Meeting_March_1st_2017
09:02:20 <oneswig> #link but really it's about planning input for the forum https://etherpad.openstack.org/p/BOS-UC-brainstorming-scientific-wg
09:04:03 <priteau> Good morning oneswig
09:04:15 <oneswig> Hi priteau, good morning
09:04:29 <oneswig> How did your chameleon review go?
09:05:09 <oneswig> It is a quiet start today...
09:05:14 <StefanPaetowJisc> Good morning folks... sorry, it seems Freenode doesn't want to play with Colloquy this morning
09:05:21 <b1airo> g'day
09:05:30 <oneswig> Hi StefanPaetowJisc b1airo
09:05:32 <oneswig> #chair b1airo
09:05:33 <openstack> Current chairs: b1airo oneswig
09:05:48 <priteau> It's been delayed… it's two weeks from now
09:05:49 <b1airo> sorry, that time of night here - have only just walked in the door. grabbing some grub...
09:06:30 <oneswig> No worries b1airo!  We were just going to look over the etherpad, which I saw you'd pre-seeded anyway
09:07:29 <oneswig> The idea of today is to compile a wish list specific to research computing use cases, and at some point I believe that will get folded in with other strands of input from other groups.
09:08:12 <oneswig> The thread from January on the operators mailing list "What would you like in Pike?" is a great start
09:09:05 <b1airo> maybe spot / preemptable instances too
09:09:37 <oneswig> b1airo: I added something on that just before the meeting, it's up there
09:09:45 <b1airo> that link is:
09:09:49 <b1airo> #link https://etherpad.openstack.org/p/BOS-UC-brainstorming-scientific-wg
09:09:52 <oneswig> StefanPaetowJisc: are you having much luck with Assent?
09:10:21 <b1airo> don't forget to name yourself on the etherpad
09:11:35 <oneswig> priteau: What thoughts on Blazar?
09:13:08 <priteau> oneswig: By the Boston summit we will have pushed out a first release (mostly bug / doc fixing), I would like to gather requirements from the community, in particular regarding interaction with pre-emptible instances
09:13:21 <verdurin> Morning.
09:13:41 <oneswig> Hi verdurin - did you see the etherpad we are on?
09:14:14 <verdurin> oneswig: looking. Can't stay long, this is my normal commute time.
09:15:41 <oneswig> verdurin: lets make the most of it then :-) what are the pain points at the Crick?
09:16:31 <priteau> oneswig: I am adding a section, just underneath pre-emptible
09:17:01 <oneswig> Saw that priteau, looks good
09:22:27 <oneswig> I'm interested in the interest in using Cinder block devices as a medium for sharing datasets.  Doesn't the nature of the device format get in the way of the data?  Or is it better that it's something that people understand, analogous to inserting the data on a CD-ROM
09:23:34 <oneswig> Either way I think it merits a bullet point all of its own
09:24:18 <b1airo> for those interested, we found a workaround for the large memory guest hugepage allocation
09:24:49 <b1airo> rather than using static hugepages we turned on THP and hard-reset the guests
09:25:38 <oneswig> I faintly recall there were issues with THP and PCI passthrough you've previously described - is that still an issue or do I have it wron?
09:27:16 <priteau> oneswig: You need to know how to mount the data in the file system (is it a block device with a MBR and one partition, or no MBR? what is the file system?), but not necessarily more complex than setting up your instances to talk with an object store (what is the path to the data? do I need credentials on my instance?)
09:28:18 <priteau> and users in the same community could settle on a set of conventions to share data this way
09:28:33 <verdurin> oneswig: yes, agree with priteau - it's a little more admin overhead but not excessive
09:28:52 <oneswig> Makes sense, simple but effective.  I hadn't thought of it like that.
09:34:17 <oneswig> On an unrelated matter, we've been making some progress with evaluating OVS offload on Mellanox NICs.  It's working on our dev system (but not in OpenStack yet).
09:34:57 <oneswig> It's not (yet) a silver bullet for VXLAN performance but it would enable us to put security groups onto our SR-IOV instances.
09:35:28 <verdurin> oneswig: interesting. Do keep us informed.
09:35:57 <mgoddard> and potentially claw back some host CPU cycles
09:36:08 <oneswig> Hi mgoddard :-)
09:36:12 <mgoddard> hi oneswig
09:36:18 <mgoddard> (and everyone else)
09:36:23 <verdurin> oneswig: have to go. Will keep an eye on the etherpad.
09:36:33 <oneswig> OK verdurin, see you
09:37:26 <b1airo> oneswig, my previous understanding was that PCI passthrough necessitated that the guest memory had to be pinned/locked and that that meant THP would not work
09:37:37 <b1airo> i think i read that on the vfio blogs
09:37:58 <b1airo> however, that does not seem to hold true in our current environment
09:38:19 <b1airo> according to AnonHugePages in /proc/meminfo
09:39:10 <oneswig> Sounds like a great subject for that recap on the hypervisor tuning guide to me!
09:42:09 <oneswig> I think we might have finished this round? Is the next step to summarise and post to the list?
09:43:28 <oneswig> Anything to add on federation capabilities?
09:45:38 <oneswig> I'm wondering if we should put anything forward on tenant logging and monitoring services
09:47:58 <StefanPaetowJisc> Logging and monitoring is always good
09:49:06 <oneswig> StefanPaetowJisc: agreed - I'm not sure what the current state of play is here (outside of our own bubble)
09:49:23 <oneswig> I'll follow up on that
09:51:08 <oneswig> OK - any final subjects for the forum for now?  We can continue via ML
09:52:05 <oneswig> #topic Any other business
09:52:22 <oneswig> Anything else to cover this week?
09:53:17 <oneswig> I'll be in the pub next Friday - might be interesting to anyone within reach of Cambridge: https://www.eventbrite.com/e/cambridge-hpc-pub-night-tickets-31441856439
09:53:29 <b1airo> etherpad is looking great
09:53:51 <StefanPaetowJisc> oneswig: I'll be in Milan... you coming?
09:54:07 <oneswig> b1airo: It's got plenty on it.  I think input from a few more people and a little refining and it'll be done.
09:54:21 <oneswig> Yes I'll be in Milan - see you there!
09:55:01 <oneswig> priteau: are you going?
09:55:20 <StefanPaetowJisc> Where are you staying? I'm at the Garden Inn.
09:55:38 <oneswig> StefanPaetowJisc: can't remember.  One of the two suggested hotels.
09:55:54 <priteau> oneswig: I am afraid not, will likely be busy preparing for the review
09:56:14 <oneswig> Ah, too bad.
09:56:37 <oneswig> Anything more?
09:57:13 <oneswig> priteau: how's Chameleon doing?
09:58:08 <priteau> oneswig: lots of planning at the moment to prepare for the call for phase 2 of the project
09:58:43 <b1airo> sounds interesting!
09:59:09 <oneswig> Anything new and exciting coming in phase 2?
09:59:38 <priteau> oneswig: Lots of it but I can't say anything about it at this point :-)
10:00:03 <oneswig> I can't wait! :-)
10:00:17 <oneswig> OK we'd better stop - time's up
10:00:31 <oneswig> Thanks all, been helpful I think
10:00:37 <StefanPaetowJisc> Alrighty!
10:00:41 <oneswig> #endmeeting