21:00:30 <oneswig> #startmeeting scientific-sig
21:00:31 <openstack> Meeting started Tue Apr  3 21:00:30 2018 UTC and is due to finish in 60 minutes.  The chair is oneswig. Information about MeetBot at http://wiki.debian.org/MeetBot.
21:00:32 <openstack> Useful Commands: #action #agreed #help #info #idea #link #topic #startvote.
21:00:34 <martial> rumble ....
21:00:35 <openstack> The meeting name has been set to 'scientific_sig'
21:00:40 <oneswig> Hello all!
21:00:43 <oneswig> #chair martial
21:00:44 <openstack> Current chairs: martial oneswig
21:01:01 <trandles> Hi there
21:01:06 <martial> hello everybody
21:01:11 <martial> Hi Tim:)
21:01:11 <oneswig> #link agenda we'll attempt to follow https://wiki.openstack.org/wiki/Scientific_SIG#IRC_Meeting_April_3rd_2018
21:01:16 <oneswig> Hey Tim
21:01:52 <oneswig> How's New Mexico?  I'm seeing a talk on Charliecloud next week I believe
21:02:04 <trandles> spring has arrived
21:02:07 <trandles> where's the talk?
21:02:19 <trandles> OH, HPCAC
21:02:19 <trandles> ?
21:02:53 <oneswig> Indeed - Michael Jennings
21:03:05 <priteau> Good evening
21:03:06 <trandles> ah yeah, Michael's a lucky guy
21:03:23 <oneswig> Le mec, glad you could make it
21:03:32 <priteau> (just one eye on IRC really)
21:03:59 <oneswig> trandles: it's a fine venue and last year's programme was pretty interesting.
21:04:31 <trandles> I'm hoping he finds time to discuss Docker & HPC with Luca from CSCS
21:05:08 <oneswig> Shifter, as in?  I believe that's what they are using.  It's Docker as a file format only, I believe.
21:05:29 <trandles> docker -> squashfs via shifter image gateway
21:05:34 <oneswig> (although that's the bit this use case requires)
21:06:00 <trandles> LANL + NERSC + CSCS had talks with Docker at SC17 about actually doing something like "runHPC" to make us all happy
21:06:30 <trandles> sorry, not trying to drag the agenda into the container gutter
21:06:34 <oneswig> I've heard it said that the new more modular OCI toolset will permit that
21:06:56 <martial> Tim: Docker Federal summit is in DC next week, I know we will not see anything out of it, but could be interesting
21:08:07 <oneswig> Government Docker?  Interesting and probably conflicted theme
21:08:14 <trandles> I <3 docker for everything up to the actual runtime...
21:08:45 <trandles> I want our code teams to use docker to build/package/distribute everything
21:08:56 <oneswig> martial: are you attending this summit?
21:09:17 <martial> Tim: https://www.docker.com/government-summit-2018
21:09:33 <martial> oneswig: will be at both the Federal summit and DockerCon18
21:09:41 <trandles> wish I could make it...but alas...
21:10:18 <trandles> oneswig: track down Christian Kniep at HPCAC...I think he's going to be there
21:10:45 * johnthetubaguy wonders in late to the back of the room
21:10:45 <oneswig> He's speaking, no less
21:10:56 <oneswig> Hey john, good evening
21:11:24 <trandles> Oh good.  I thought Christian might have been speaking about HPC and Docker at the Gov't Summit.  Better that he be at HPCAC IMO
21:11:30 <oneswig> trandles: is he from LANL?
21:11:38 <trandles> Nope, he works for Docker
21:12:09 <oneswig> Ah, so he does.  I mangled the talk title with the vendor
21:12:14 <trandles> he's the contact we've (LANL/NERSC/CSCS) been working with
21:12:47 <oneswig> I'll be doubly interested in what he has to say then.
21:13:14 <trandles> Michael was also in the meeting at SC17
21:13:28 <trandles> so maybe you three could have a chat
21:14:18 <oneswig> I'll add it to my list - I'm already looking forward to several useful catch-ups.
21:14:28 <oneswig> Not least with DK Panda on RDMA and Hbase
21:14:59 <oneswig> We should probably get through the agenda and pick this up in AOB...
21:15:28 <oneswig> #topic SIG RFEs for Ironic
21:16:00 <oneswig> We had some good discussions at the PTG
21:16:18 <oneswig> And the conclusion was that those discussions should be captured in storyboard
21:16:51 <oneswig> The two items were boot-to-ramdisk and boot-with-kexec
21:17:33 <oneswig> but right now they are sitting idle waiting for specs to elaborate on the idea.
21:17:59 <oneswig> I think if we truly want these things, we'll need to articulate our needs :-)
21:18:05 <trandles> I was just reading over the items that Julia linked on the boot-to-ramdisk storyboard
21:19:00 <oneswig> The specs process wiki page?
21:19:12 <rbudden> I believe I still need to add the Bridges use case to the spec/blueprint/etc
21:19:15 <trandles> that and the template
21:19:36 <rbudden> Are the use cases what we still need?
21:19:39 <trandles> to me the template implies a lot more depth of knowledge of the inner workings of ironic
21:19:47 <trandles> more than I might possess
21:20:03 <trandles> #link https://github.com/openstack/ironic-specs/blob/master/specs/template.rst
21:20:55 <oneswig> This may end up as a team effort
21:20:58 <johnthetubaguy> not sure ironic does this, but Nova tried out backlog specs, where you only fill in the bits you know about
21:21:05 <johnthetubaguy> that is a good way to get started at least
21:21:30 <oneswig> johnthetubaguy: interesting.  But referred to in the past tense - what happened?
21:21:53 <priteau> trandles: you could mark some section as unknown and ask during the review if an Ironic contributor can helps define them
21:22:26 <johnthetubaguy> oneswig: more it didn't really help things gain traction https://specs.openstack.org/openstack/nova-specs/specs/backlog/index.html
21:23:24 <oneswig> I picked the parallel filter scheduler at random, it's still highly detailed
21:23:48 <johnthetubaguy> that might be my fault... as I wrote that one
21:24:08 <oneswig> ha!
21:24:08 <johnthetubaguy> FWIW, placement implements that, largely
21:24:19 <trandles> flavor classes one looks a lot more like what I'd produce at this point
21:24:30 <johnthetubaguy> yeah, that's cool
21:24:49 <johnthetubaguy> its probably best having a change in gerrit that someone else can help finish off
21:25:00 <johnthetubaguy> writing it down often proved interesting :)
21:25:14 <johnthetubaguy> s/proved/proves/
21:25:29 <oneswig> Given the flavor classes spec was proposed for Liberty, I take johnthetubaguy's point on traction
21:26:00 <martial> :plus_one: :)
21:26:56 <oneswig> It seems the best plan may be to work on embellishing a spec between us
21:27:00 <johnthetubaguy> turns out if no one implements a thing it doesn't happen
21:27:14 <trandles> oneswig: agreed
21:27:21 <oneswig> trandles: if you were to lay down the skeleton, and add us all as reviewers?
21:27:33 <trandles> sure, I can get started on it later this week
21:28:03 <oneswig> Is this for boot-to-ramdisk or boot-to-kexec? Or both?
21:28:20 <martial> sounds like a good plan actually to get work done
21:28:56 <trandles> ramdisk for me
21:29:05 <oneswig> martial: steady on..
21:29:13 <trandles> priteau: I believe kexec was your baby
21:29:18 <priteau> Yep
21:29:51 <oneswig> ... and rbudden had the epic bios cycle times driving that
21:30:02 <rbudden> yes
21:30:08 <priteau> I can write a spec as well
21:30:28 <rbudden> boot to ramdisk would be nice for the Superdomes :)
21:30:44 <rbudden> one boot, entire OS in RAM
21:31:07 <rbudden> kexec would be nice though for speeding up normal nodes, so i’m in for that as well
21:31:10 <rbudden> let me know what I can do
21:31:31 <oneswig> Great, thanks Bob and thanks all
21:31:33 <trandles> then again we should all be using linuxboot.org in a few months...or so another crowd has been telling me...
21:31:49 <oneswig> not come across that...
21:32:30 <oneswig> Reminds me of Ron Minnich, of your parish trandles
21:32:38 <trandles> it's a tangent, but search Ron Minnich
21:32:40 <trandles> yeah, it's him
21:33:02 <trandles> although he's at google these days
21:33:21 <oneswig> Fancy that.  I came across him via the Alpha CPU and Linux BIOS
21:34:22 <oneswig> OK, interesting.  Lets move on?
21:34:33 <oneswig> #topic Vancouver
21:34:48 <martial> last day to get the early bird ...
21:35:28 <martial> #link https://www.openstack.org/summit/vancouver-2018/
21:35:34 <oneswig> My first trip to this great city.  Already looking forward to it.
21:35:51 <martial> the last summit there was really good
21:36:14 <johnthetubaguy> +1 it was good last time, nice ducks on the roof outside the breakout rooms
21:36:23 <rbudden> +1
21:36:30 <johnthetubaguy> I don't think I am making that up
21:36:32 <rbudden> will be nice to revisit Vancouver
21:36:39 <oneswig> We should ensure the Scientific SIG sessions are truly awesome... any thoughts?
21:36:45 <trandles> plus watching the float planes...unless it rains this time...
21:37:08 <b1airo> Vancouver gets high marks for beer quality too
21:37:19 <martial> wait we have a conference around watching the duck and planes ... and beer
21:37:22 <johnthetubaguy> I am cooking some TC related things for a Scientific Constellation, to sit alongside a container one
21:37:25 <oneswig> Hey b1airo
21:37:34 <martial> #chair b1airo
21:37:35 <openstack> Current chairs: b1airo martial oneswig
21:37:45 <martial> welcome Blair
21:37:50 <johnthetubaguy> it would be good to do something in person around that, hoping to get something up for review soon ish
21:38:04 <oneswig> johnthetubaguy: can you give an overview of the concept?
21:38:07 <b1airo> o/ (have been lurking)
21:38:11 <martial> #link Forum brainstorming https://etherpad.openstack.org/p/YVR18-scientific-sig-brainstorming
21:38:16 <johnthetubaguy> #link https://governance.openstack.org/tc/resolutions/20170404-vision-2019.html#navigating-with-constellations
21:38:26 <johnthetubaguy> so thats the context
21:38:27 <martial> (oops too soon)
21:38:50 <johnthetubaguy> ah, this is forum brainstorming too really
21:39:30 <oneswig> This is interesting: "Many of the tools used for OpenStack deployment now provide high level macros to install specific constellations"
21:39:31 <johnthetubaguy> basically, its a bit like a reference architecture, or a way to navigate through the huge list of openstack projects but looking for your use case, rather than "insert crazy name"
21:39:47 <johnthetubaguy> we got kinda inventive when we wrote that vision
21:39:53 <oneswig> which tools are they johnthetubaguy?  Perhaps we need a "kayobe install ska"?
21:40:07 <johnthetubaguy> I am not against that idea, obviously ;)
21:40:30 <johnthetubaguy> or maybe just "vagrant up"
21:40:33 <oneswig> Everyone will want one :-)
21:41:44 <johnthetubaguy> first step is describe what it looks like, roughly, and start making the concept a bit more real
21:41:53 <johnthetubaguy> so will try keep folks posted on how that goes
21:42:27 <johnthetubaguy> I am hoping dims does an awesome job of the container one so it makes it easier :)
21:42:38 <oneswig> Is there a forum-style discussion around this or is it something that's just going to work its way through OpenStack's processes?
21:43:25 <oneswig> I saw that noggin143 has added a forum proposal on preemptible instances - great
21:44:07 <johnthetubaguy> so I am not sure yet... I was thinking best to discuss in the Scientific SIG room
21:44:28 <b1airo> Sounds good
21:44:52 <johnthetubaguy> preemptible should be a good session too, good progress happening there
21:45:18 <oneswig> excellent.  I think it will be disruptive
21:45:28 <oneswig> (and not just for my instances)
21:45:54 <b1airo> I suspect the idea of Scientific Constellation will flush out a few snowflakes
21:47:19 <oneswig> johnthetubaguy: is there a forum topic in the federation experiences I wonder?
21:48:00 <johnthetubaguy> very possibly
21:48:09 <johnthetubaguy> at least worth a sync up
21:48:11 <b1airo> Also wonder how much a Science community specific stuff there might be in config, e.g., Keystone setup and various API Policy
21:48:34 <b1airo> s/ a//
21:51:42 <oneswig> A further thought on Vancouver - is there somewhere where might all have a beer and watch sea planes, ducks, etc? We should probably (informally) make a plan
21:52:06 <rbudden> there’s a decent brewery within walking distance of the conference center
21:52:24 <rbudden> I forget the name but it’s close to the famous Vancouver steam clock
21:52:37 <oneswig> Great plan
21:52:46 <trandles> isn't there a bar/restaurant right next to the seaplane base and the conference center?
21:52:55 <trandles> although IIRC it was busy and expensive
21:52:59 <b1airo> I'm arriving Friday evening so plenty of time
21:53:33 <oneswig> b1airo: is it direct for you?  That's a lot of ocean
21:53:45 <martial> b1airo are you participating in the upstream academy?
21:54:27 <b1airo> Via Auckland
21:55:06 <oneswig> Since when was New Zealand on the way to anywhere...
21:55:12 <oneswig> 5 mins to go
21:55:17 <oneswig> #topic AOB
21:55:40 <oneswig> What's new - got your DGX2 orders in
21:56:06 <b1airo> Ha. If our grant gets funded then yes
21:56:41 <oneswig> I've been tinkering with RDMA for Ceph.  Sometimes it's working but IB verbs portability is low it seems.
21:57:02 <b1airo> Between fabric types you mean?
21:57:09 <oneswig> I can guess it's being developed on RoCE because that works.
21:57:42 <oneswig> Nothing else works out of the box.  I'm teaching myself verbs to get it coaxed into life
21:58:27 <b1airo> I seem to recall a discussion on ceph-users a while back about connection setup. How does Ceph initialise QPs?
21:58:51 <oneswig> Directly. ibv_modify_qp.  Really should use CMA
21:59:08 <b1airo> Hmm yes
21:59:31 <martial> (running out of time, gents)
21:59:33 <b1airo> Sounds like it's still experimental at best
21:59:34 <oneswig> The Ceph messenger classes have enough flexibility to support that.
22:00:00 <oneswig> b1airo: agreed, but there is potential here.  For the day when the CPU is the bottleneck for Ceph.
22:00:15 <oneswig> Ah, on that happy note, we must close
22:00:20 <oneswig> thanks all
22:00:24 <oneswig> #endmeeting