21:00:44 <oneswig> #startmeeting scientific-sig
21:00:45 <openstack> Meeting started Tue May 12 21:00:44 2020 UTC and is due to finish in 60 minutes.  The chair is oneswig. Information about MeetBot at http://wiki.debian.org/MeetBot.
21:00:46 <openstack> Useful Commands: #action #agreed #help #info #idea #link #topic #startvote.
21:00:48 <openstack> The meeting name has been set to 'scientific_sig'
21:01:16 <oneswig> #chair b1airo martial
21:01:17 <openstack> Current chairs: b1airo martial oneswig
21:01:21 <oneswig> hey guys
21:01:26 <b1airo> o/
21:01:35 <oneswig> How's things?
21:02:13 <jmlowe> Doing ok myself, you?
21:02:38 <oneswig> having a good one today.  Performance analysis, when it yields something, is fun.
21:03:18 <martial> looking into gitlab for swagger/open api usage ... maybe to create a new project for the P2302 effort
21:03:36 <oneswig> As in the NIST cloud federation work?
21:04:17 <jmlowe> I believe I had a bit of a breakthrough today
21:05:31 <jmlowe> Was struggling to come up with a way to effectively use all the bandwidth to cephfs, I think using a combination of DVR, BGP dynamic routing, and IPv6 gets me there
21:06:02 <martial> well the CFRA is out, NIST SP500-332
21:06:07 <oneswig> Does IPV6 work with the other two jmlowe?
21:06:13 <martial> right now we are looking at the IEE side
21:06:20 <jmlowe> Yes, I believe it does
21:06:30 <oneswig> Just checking :-)
21:06:33 <martial> and we want to define a minimalistic API definition
21:06:38 <martial> so OpenAPI compatible
21:06:48 <martial> and open to all to contribute
21:07:09 <jmlowe> added bonus is that floating ip address get the full compute node bandwidth and can bypass the network node
21:07:09 <martial> so was thinking gitlab since they have an OpenAPI viewer
21:07:17 <jmlowe> for ipv4
21:07:54 <martial> (brb)
21:08:29 <oneswig> jmlowe: weren't you using linuxbridge on Jetstream or is this a new project?
21:09:36 <jmlowe> yes we are using linuxbridge, and dvr now works with linuxbridge, I'm thinking about ways to reinvent Jetstream
21:10:48 <trandles> Sorry I'm late. Prior call ran long.
21:10:57 <oneswig> Sounds good jmlowe.  I'm wary of DVR but perhaps that's undue
21:11:05 <rbudden> ditto, hello all
21:11:14 <oneswig> hi trandles rbudden, glad you could make it :-)
21:11:22 <jmlowe> rbudden do you guys use dvr with linuxbridge?
21:11:26 <rbudden> yes
21:11:34 <rbudden> dvr_noext
21:12:10 <rbudden> since we’re mostly provider networks in our one cloud
21:12:48 <jmlowe> With linuxbridge it's relatively simple, proxy arp and dnat in the network node network ns, compute node gets snat rules and dumps things out onto the provider network
21:14:32 <jmlowe> a bit of asymetric routing, throw in bgp dynamic routing and you can move the dnat to the compute node
21:14:48 <oneswig> In a parallel universe, I've been learning much about OVS hardware offloads and how to understand their dark magic.
21:15:26 <jmlowe> Anything useful to report from your trips to that netherworld?
21:15:42 <oneswig> Yes.  Don't ever go there, unless performance makes you do it :-)
21:17:28 <jmlowe> I've been going the opposite direction, looking at BGP as an elegant weapon for a more civilized age
21:17:37 <oneswig> Using SRIOV I'm getting line rate from the 100G NIC, but currently only one of a bonded pair attached to my OVS bridge.  OVS flow rules imply I should be using both bonded links, active-active.  The silicon's not quite on the same page, but I am so close...
21:18:12 <oneswig> jmlowe: BGP to the hypervisor - cool stuff.  What do you need to do to make it work?
21:20:15 <jmlowe> afaik just the bgp speaker agent from neutron
21:21:02 <oneswig> What does the network fabric look like?
21:22:56 <jmlowe> Well, that seems to be left as an exercise for the user
21:23:45 <oneswig> ... you get your users to cable up their own networks? :-)
21:24:48 <jmlowe> I think you can use frrouting and vxlan to publish routes to the computes and keep the tenants non the wiser
21:26:47 <oneswig> I'd be very interested to hear how you get on with that.  The layer-2 equivalent will look like a poor country cousin once you're done
21:28:19 <oneswig> I'm having an interesting MLAG issue currently, trying to get ingress traffic to use both ports instead of 95% of it coming to one port.
21:32:51 <martial> SYN
21:33:03 <jmlowe> which lacp mode are you using?
21:33:51 <oneswig> lacp layer 3+4 - should be doing the right thing.  I wonder if the hardware hashing isn't including source IP and MAC, that might be the only uniqueness between flows.
21:37:57 <oneswig> I should check, does anyone have any points to raise - agenda items, workshops, etc?
21:43:18 <martial> Tim?
21:43:37 <jmlowe> Not I, still holding out hope the Berlin Summit might happen
21:43:51 <martial> same
21:44:01 <rbudden> jmlowe: +1
21:44:52 <oneswig> For the ex-Vancouver PTG, we should think of those upstream activities we always wanted to do, but never had the time.
21:45:01 <trandles> I'd say something about the DOE Workflow summer seminar series but the schedule hasn't been posted yet.
21:45:17 <martial> so Stig, you are looking to do RTX 2080 passthrough?
21:45:29 <trandles> Eventually at https://wowoha.org
21:45:57 <oneswig> martial: asking for a friend, really.
21:46:31 <oneswig> Coincidentally I think we'd just did the setup to get them working today.
21:46:40 <martial> oh that's cool
21:46:53 <trandles> Big doubts that Berlin happens. Why SC20 hasn't gone virtual yet I don't know. There's about 0.1% chance I'm going anywhere near Georgia before I can get a vaccine. Plus, with COVID-19, it's worse now.
21:48:18 <martial> this is a little sad in truth
21:48:33 <jmlowe> I don't see how they can actually pull of a virtual SC20
21:49:15 <oneswig> I don't think they even put the keynotes online from SC19 did they?  I remember looking for them some weeks later and didn't find it.
21:50:04 <trandles> Nope and I even emailed the organizers asking about that. They said to keep checking back but the videos would be posted. I never found them.
21:50:30 <b1airo> lol trandles !
21:50:51 <oneswig> perhaps someone forgot to hit record on the day :-)
21:51:03 <martial> technology is hard
21:51:25 <oneswig> Anyway, SC has some work to do to catch up with the online world.
21:53:02 <trandles> It'll be interesting to see how ISC20 does online
21:53:24 <trandles> I heard mixed reviews for Red Hat's summit
21:53:28 <b1airo> an unrelated question - anyone claim to be a slurm aficionado? i'm investigating consistently high "reserved time" and wondering if it is something to do with all the array jobs on this machine
21:53:36 <trandles> Sounds like the system was overwhelmed somewhat
21:54:28 <oneswig> b1airo: not an aficionado to that degree, sorry
21:54:39 <trandles> b1airo, I wouldn't saw aficionado but I can put you in touch with our slurm gurus
21:56:09 <b1airo> thanks trandles - i might reach out if i don't hear back from my usual suspects (one local and one at NERSC). on the other hand I could probably just submit a support ticket and then blog about it
21:59:56 <oneswig> Nearly at the hour - anything more to add?
22:00:41 <oneswig> OK y'all, have a good day
22:00:45 <oneswig> #endmeeting