21:01:49 #startmeeting scientific-wg 21:01:50 Meeting started Tue Sep 6 21:01:49 2016 UTC and is due to finish in 60 minutes. The chair is b1airo. Information about MeetBot at http://wiki.debian.org/MeetBot. 21:01:52 Useful Commands: #action #agreed #help #info #idea #link #topic #startvote. 21:01:54 The meeting name has been set to 'scientific_wg' 21:02:03 * anteaya pictures b1airo as Mr Music 21:02:15 an easy picture! 21:02:23 Good evening everyone 21:02:31 heh, you happen to have that agenda handy oneswig ? 21:02:33 Hello 21:02:38 coming right up 21:02:50 #link agenda for today's meeting https://wiki.openstack.org/wiki/Scientific_working_group#IRC_Meeting_September_6th_2016 21:03:45 and inline... 21:03:47 WG activity space requested at the Barcelona summit. 21:03:47 BoF for poster sessions and lightning talks 21:03:48 Working group meeting for progress review and planning 21:03:48 Possible venue for social (Thursday night) 21:03:48 Supercomputing preparation 21:03:48 Discussion topics for SC panel session 21:03:50 Priority items for SC BoF (which was accepted!) 21:03:52 Other business 21:03:54 Diary date - OpenStack/HPC meetup in London on 1st December 21:03:56 Cloud-native applications for research computing - target list for App Ecosystem WG 21:04:00 hmm sorry about the formatting 21:04:11 Ah, forgot you preferred it that way 21:04:17 #topic WG activity space requested at the Barcelona summit 21:04:38 hello 21:04:49 I put in for two sessions but perhaps we should request another based on user-committee mails today 21:04:57 Hi rbudden! 21:05:04 sorry, i’m a tad late! 21:05:11 o/ 21:05:32 Question was whether we could get a BoF session together, and whether it would overspill one room into two 21:05:40 No use if they are not close by though 21:05:57 yeah 21:06:29 i'm not overly hopeful, sounds like things are gonna be tight 21:06:44 This session 2 weeks ago, martial suggested it might be good to bring posters - if anyone has one handy? 21:07:12 blairo: it does. I'll ask anyway, we have 2 sessions reserved. 21:07:43 i for one do not have any poster fodder readily available 21:08:10 but i suppose there is a chance we could create something as an artifact of our joint summit presentation... 21:08:41 (...need to ping wojciech about that) 21:08:47 Me neither but that's a good idea 21:09:13 Wojciech has tonsilitis, poor guy, will be offline for a few days 21:09:33 nasty 21:09:44 So don't expect an immediate response 21:09:55 i certainly like the idea of lightening talks in a scientific-wg bof 21:10:09 But a poster from the Lustre talk is a good idea, and talks are also good 21:10:57 anybody present keen to talk about their recent challenges and/or wins related to openstack for science/hpc ? 21:11:16 (in a lightening talk in barcelona) 21:11:38 maybe, I'll have to think of something 21:11:48 I'm happy to talk on either getting SRIOV devices configured in a TripleO deployment or configuring Dell BIOS using Ansible 21:12:00 ... been one crazy summer 21:12:31 Note as actions perhaps? 21:12:36 (sorry I am late) 21:12:39 I might have something in lightening form to say about deploying 300 nodes using salt 21:12:57 was thinking that but maybe let's hold it till a later date when we have more concrete details of summit sessions ? 21:13:12 hi martial ! 21:13:23 jmlowe, that sounds pretty interesting 21:13:26 BM nodes? 21:13:31 I could give a quick presentation of the Blazar project and how we use it 21:14:02 priteau, that would definitely be nice, i suspect a bunch of folks don't even know what it does 21:14:19 priteau: I'd go to that one 21:14:58 priteau: saw your UI today on chameleoncloud.org, looks good, I had no idea Blazar had that 21:16:07 the Horizon support for Blazar that we have is half integration of two year old patches never merged, half own development 21:16:38 Patches that old, do they become like fine wine? 21:17:04 damn, australian access federation doesn't get me into chameleon 21:17:40 or mouldy cheese perhaps ? :-D 21:17:56 OK it sounds like good enthusiasm for a lightning talk model right now, lets go with that and fill in the slots nearer the time 21:18:09 oneswig: some age better than others ;-) 21:18:23 I may be able to do a lightning talk about using Ansible in conjunction with the Slurm prologue to allow Slurm reservations to automate the setup of Nova Compute/OVS for reserving traditional cluster nodes to host VMs/Containers 21:18:44 b1airo: we're only federated with GENI so far 21:19:06 rbudden: sounds good, what's the impact on job startup time? 21:19:13 b1airo: seems like AU/EU/US NSF federation should be a thing, especially with cloud resources 21:19:16 rbudden, a sort of dynamic slurm? that sounds neat 21:19:23 rbudden: I would be very interested to hear about this 21:20:20 All the more reason to attend the WG meeting then... 21:20:43 we use slurm at monash but i'm not close enough to hpc queue admin-ing to have looked into existing slurm support for on-demand expansion/contraction of the cluster 21:21:17 is there such a thing already, or just the hooks? 21:21:23 this is all in the early testing phases but we have a few non traditional use cases where we’ve had to spin up large quantities of VMs 21:22:14 the idea was to have Slurm feed Ansible the requirements so we could metadata tag the nova hypervisors and auto create host aggregates for the user on demand 21:22:35 The people at OSU use SLURM in a similar way IIRC, VMs are created for running each workload. Seems like hard work for the control plane if workloads are short lived? 21:22:57 I think they do it to inject the inter-VM SHMEM gadget 21:22:59 that way they could do a slurm reservation and within a few min have the resources just show up inside the dashboard and have proper flavors associated with the nodes 21:23:49 rbudden: so you salloc in slurm and nova instances are created? Ooh. 21:23:55 rbudden, ahh with you now, interesting, catering to "i want my own environment / i need root" hpc users 21:24:04 yep, exactly 21:24:36 we use slurm for all our accounting, so it makes it easy to charge since the reservation is tied to a project/charge-id 21:24:49 i would be very interested to see how that turns out, we have some of those cases here too 21:25:11 grumble grumble bioinformaticians 21:25:52 How does slurm know how many resources are available to it, and what happens when someone else uses them? 21:26:03 of course, giving them a chunk of a parallel filesystem securely is another challenge 21:26:56 we have slurm tied in with pam, so only admins and users who have an active job/servervation can login to the nodes they have been given 21:27:36 b1airo: yes, filesystem access is complicated 21:28:04 the majority of our VMs right now are gateway type workloads and they want access to distributed filesystems of Bridges 21:28:10 so they don’t get root on those ;) 21:28:34 We probably ought to move on? 21:28:35 fair enough! 21:28:40 eventually they can on our Slash2 filesystem because i’ll be implementing ticket based auth 21:29:09 so that trust is not at the client level 21:29:23 rbudden, how modern! :-) 21:29:32 feel free to ping offline if ppl want to banter about it ;) 21:30:13 rbudden: might just do that 21:30:20 :) 21:30:25 yeah this sort of area would be good delve into more detail on 21:30:49 #action: take PFS auth discussion to mailing list 21:31:22 #help 21:31:30 Is anyone hungry? 21:31:42 hmm meetbot may have gone awry 21:31:55 yeah, last dot point here was drinks/dinner in barcelona 21:32:12 #link Possible candidate for Thursday evening in Barcelona - after Blair's talk - http://www.asadordearanda.net/en 21:32:33 recommended by a former workmate who is an occasion local 21:33:05 i’m hungry now 21:33:25 It's under 5km from the conference venue, 8 stops on the metro. Could fit the bill 21:33:39 I checked today, they have room for 30 :-) 21:33:40 I do have one concern, I'm not planning on bringing much if any cash, so if they are more friendly to a big pile of credit cards than the place in Austin was that would be good 21:34:53 jmlowe: hard to know, European places might be more open to it, cash is less prevalent than card AFAIK 21:35:37 they have a bunch of set group menu options for ~45 euro / head 21:36:09 Yes, thats' for something pretty big, I was reckoning on a course from the basic menu and some beer or wine... 21:36:47 all of it sounds very nice so far :) 21:36:55 we did that at a Lebanese restaurant with a smaller group in Vancouver, it was great 21:37:08 oneswig: They have three venus in Barcelona, which one did you have in mind? 21:37:10 jmlowe: sounds ideal. 21:37:18 Calle ...err ... holdon 21:37:25 jmlowe: yes, that was fun indeed 21:37:38 Pau Claris 21:37:58 Vancouver was so far the winning summit for beer 21:38:06 IMHO 21:38:16 they did have good beer 21:38:19 gotta get the basics right! 21:38:21 gelatto was amazing too 21:39:35 http://www.asadordearanda.net/en/content/barcelona-pau-claris 21:40:05 Nearer the time, I might need to get a head count to ensure the booking's within a power of 10 21:40:13 lol 21:40:34 i assume it's just like Ceph PGs, round up to the nearest power of 2 ? 21:40:49 Tell that to the maitre d', he'll love it 21:41:08 i'd prefer to leave with my teeth intact 21:41:20 lol indeed! 21:41:29 #topic Supercomputing preparation 21:42:01 firstly, our BOF was accepted - w00t! 21:42:09 excellent! 21:42:11 nice 21:42:17 Does anyone know which days the panel and bof are? 21:42:27 Pannel is Thursday? 21:42:42 panel is thursday at 3:30-5pm 21:42:47 jmlowe: do you have a link to it? 21:42:49 yeah that's right for panel, not sure the BOF timing is set yet though 21:43:03 checking.... 21:43:09 http://sc16.supercomputing.org/presentation/?id=pan127&sess=sess184 21:43:27 really sweet on the BOF :) 21:43:56 http://sc16.supercomputing.org/presentation/?id=bof180&sess=sess361 21:44:34 ah there it is! 21:44:50 Wonderful, thanks jmlowe rbudden 21:47:13 i thought we take a moment to brain-dump some discussion points for the SC panel that might help get things underway 21:48:09 It's an interesting one, who's a typical attendee? Cloud skeptics will probably steer clear. 21:50:00 In writing this study, I've learned some interesting capabilities I didn't know about what can be achieved in OpenStack. Do you think this is an infrastructure-level discussion or higher level? 21:51:30 i think if we get it right it will range higher level down to infrastructure and back to "business" level again 21:52:16 #link SC16 panel brainstorm etherpad - https://etherpad.openstack.org/p/scientificwg-SC16-panel 21:53:13 Good plan, I'll keep thinking on it in the background 21:54:29 excellent 21:54:30 Next item? 21:55:31 #topic AOB 21:55:52 london meetup oneswig ? 21:56:09 I had one item - limited use in this group - London meetup for HPC / OpenStack is now 1st December 21:56:26 Sponsored by Mellanox 21:56:43 we're having an OpenStack Australia Government Day here soon - http://australiaday.openstack.org.au/ 21:57:42 This is selling to Government or government of OpenStack?? 21:57:49 haha 21:58:11 yeah i think it's more aptira selling to government 21:58:22 or hoping to anyway 21:59:01 Turns out the UK gov likes OpenStack, why not AU... 22:00:09 i have no idea where gov is at with private/hybrid cloud, but to complicate matters we have federal and state govs with separate practices and departmental guidelines 22:00:34 OK we are out of time alas, 22:00:46 I am left wondering why they chose an image with no clouds in it... 22:00:58 little bit silly with states of ~6-7 million ppl at most 22:01:06 So typical of Australia 22:01:09 :-) 22:01:32 i will turn to my NZ citizenship at this point 22:01:37 ;-) 22:01:41 Got to go, thanks all 22:01:52 pleasure as always 22:01:52 thanks 22:01:57 thanks folks! 22:02:07 #endmeeting