21:01:46 <martial__> #startmeeting Scientific-sig
21:01:47 <openstack> Meeting started Tue Dec 11 21:01:46 2018 UTC and is due to finish in 60 minutes.  The chair is martial__. Information about MeetBot at http://wiki.debian.org/MeetBot.
21:01:48 <openstack> Useful Commands: #action #agreed #help #info #idea #link #topic #startvote.
21:01:50 <openstack> The meeting name has been set to 'scientific_sig'
21:02:51 <martial__> Opening the floor
21:04:02 <janders> who's around today? It seems like things are getting quiet moving towards xmas..
21:04:16 <martial__> it is indeed usually the case
21:04:38 <martial__> Stig and Blair are not able to join, so I offered to run an open meeting
21:04:47 <janders> great! :)
21:05:02 <martial__> if there is a topic you want to cover or info you want to share, please let us know
21:05:03 <janders> I don't have a whole lot of stuff to update on - working on training & procurement
21:05:25 <janders> one thing I've been thinking of / looking into recently is BeeGFS - OpenStack integration
21:06:15 <janders> it seems to me it's a very promising area, but with little work done so far
21:06:27 <janders> do you have any experience or thoughts on this?
21:07:04 <martial__> I know this has been discussed recently in the channel
21:07:15 <martial__> let me find the link
21:07:51 <martial__> BeeGFS [...] met them at SC18. Them and OpenIP seemed interesting to look into, so putting it out here https://www.enterprisetech.com/2018/11/29/move-over-lustre-spectrum-scale-here-comes-beegfs/
21:08:25 <janders> that's right - we (CSIRO) even got a mention :)
21:08:36 <martial__> not sure if this link was taken from a Scientific SIG conversation
21:08:48 <martial__> looks like it was from here then :)
21:09:47 <martial__> not much more from my end on this subject yet
21:10:02 <janders> would you see value in going down this path in the future?
21:10:29 <janders> from my perspective, BeeGFS is set to be Ceph equivalent for High Performance Cloud workloads
21:10:56 <janders> it's an opportunity, but whether it translates into success depends on thinkparq's interest and update from people like us
21:11:32 <janders> s/update/uptake
21:11:51 <martial__> not sure, Stig mentioned something in ... October about doing some test
21:12:23 <martial__> Oct 2nd log `We did a little testing on resource footprint with hyperconverged (non-containerised) on bare metal.  I think the containers could work.` `On the extreme bandwidth topic, one of our team has been getting really great numbers from BeeGFS at scale recently.  Servers with NVME and dual OPA exceeding 21GBytes/s for reads.  Yeee-ha`
21:12:43 <janders> regarding our BeeGFS work - I can't say too much yet (will be happy to share when it's complete) but we're somewhere around 80% efficiency in terms of BeeGFS being able to utilise the performance of the underlying hardware
21:13:09 <janders> I think we're roughly around the same mark - which is not unexpected given we're using very similar hardware
21:14:26 <martial__> seems pretty good indeed
21:14:52 <janders> the guys in Cambridge are doing excellent work trailblazing on the topic of ultra-high-performance BeeGFS, we're following closely here at CSIRO :)
21:15:18 <janders> I believe there'll be Cambridge-DELL whitepaper on the topic out soon, looking forward to reading it
21:15:49 <janders> ok - so that's really the only thing I had - other than BeeGFS is all training and procurement
21:16:26 <janders> 80% of my team is away doing a mix of different courses, so not much engineering happening
21:17:03 <janders> how are things on your end Martial?
21:19:17 <martial__> doing good, working on a K8s deplpoyment over kubespray on top of OpenStack kolla-ansible
21:20:58 <janders> I don't think I've looked at kubespray before - quickly checked it out, looks very cool
21:21:09 <janders> if it's k8s + ansible I already like it :)
21:21:10 <martial__> very simple to deploy a k8s
21:21:34 <martial__> so if you have an OpenStack project you have a private network
21:21:43 <martial__> and you just deploy a private k8s on that private network
21:22:09 <martial__> very cool for quick testing
21:22:15 <martial__> and prototyping
21:22:17 <janders> excellent!
21:22:35 <martial__> (and yes there is an kubernetes-ansible :) )
21:22:41 <janders> can it be made work for larger scale production deployments, or is it focusing specifically on PoC/prototyping?
21:22:56 <martial__> you can make large deployments with it
21:23:26 <janders> https://github.com/kubernetes/contrib/tree/master/ansible
21:23:26 <martial__> OpenStack-Kolla+O-K-Ansible is how we are starting to move our OpenStack deplopyments
21:23:37 <janders> Kubernetes Ansible DEPRECATED: See repo https://github.com/kubernetes-incubator/kubespray
21:23:50 <janders> am I looking at the right repo?
21:24:04 <janders> things are moving so quickly in this field it's sometimes hard to keep track
21:24:29 <martial__> weird, we are using it in prod
21:24:52 <janders> from this it seems k8s-ansible turned into kubespray - does this sound possible, or am I getting confused? :)
21:24:53 <martial__> check https://github.com/kubernetes-sigs/kubespray
21:26:02 <janders> is openstack-kolla using lxc or docker for the control plane containers?
21:26:12 <martial__> docker
21:26:41 <janders> great!
21:27:00 <janders> are openstack-kolla and kolla-ansible very different?
21:27:52 <janders> or are these different pieces to the same puzzle?
21:31:12 <martial__> kolla creates the containers
21:31:23 <martial__> kolla-ansible is for deployment of said containers
21:31:45 <martial__> (sorry a couple people in the room asking me questions)
21:33:28 <janders> ok - so same pieces to the puzzle :)
21:33:45 <janders> are you running your control plane containers on docker/standalone or k8s?
21:34:29 <janders> we'll likely be standalone in the first iteration of the production system, but it's an interesting one - I wonder how others do it
21:35:04 <martial__> we do docker standalone
21:35:31 <martial__> once I have a project, I deploy a k8s using kubespray on created VMs
21:36:01 <janders> do you reqiure multi-tenancy in the k8s layer?
21:36:18 <martial__> not for the experiment I am running
21:36:22 <martial__> one project, one user
21:37:15 <janders> understood
21:37:46 <janders> I had some interesting discussions with Red Hat guys in Berlin about multi-tenant k8s/ocp on OpenStack which I need to follow up
21:38:14 <janders> Rocky brings improvements in this field but IMHO it's still lacking a couple features before being ready to be used in anger
21:38:54 <janders> Kuryr seems to be finally doing something useful, but it's lacking mapping of OpenStack tenants to k8s tenants AFAIK
21:39:16 <janders> so I think taking a focused approach like yours has potential of saving quite a few headaches :)
21:39:56 <martial__> works for my tailored need indeed :)
21:41:44 <martial__> seems that we managed to talk about a few topics here today after all :)
21:41:53 <martial__> anything else to cover?
21:41:54 <janders> indeed! :)
21:42:01 <janders> no, not really
21:42:14 <janders> I've got a meeting at 900AEDT so happy to wrap up early
21:42:14 <martial__> otherwise I invite us to save 20 minutes ;)
21:42:25 <janders> thank you Martial!
21:42:27 <janders> have a good day mate
21:42:34 <martial__> thanks Janders
21:42:37 <martial__> you as well
21:42:40 <martial__> #endmeeting