#opendaylight-integration: s3p
Meeting started by dfarrell07 at 15:38:51 UTC
(full logs).
Meeting summary
-
- dfarrell07 mgkwill Luis Sanjib Vratko and Al
are here (dfarrell07,
15:39:41)
- about 2 weeks until release, so have that long
to get test code in place and ready for testing (dfarrell07,
15:41:17)
- Luis has been making good progress on OF
tests (dfarrell07,
15:41:26)
- Luis has been working on bulk-o-matic in
sandbox, NB test next (details in CPerf meeting logs), latency test
on SB next, then scale, then longevity (will likely not have time
for all, but that's the prio list) (dfarrell07,
15:42:20)
- Sanjib has a 3-node cluster job pushed, but
hitting issues and needs some refacotring, so still working on
single node cluster for now (dfarrell07,
15:43:10)
- Luis demos job in sandbox bulkomatic flow add
rate (dfarrell07,
15:44:19)
- plotting confg ds, op ds, and in-switch
times (dfarrell07,
15:44:45)
- running these tests for both Be and B
(dfarrell07,
15:45:13)
- B takes a bit more time to config the switch,
but stats come much faster (dfarrell07,
15:45:35)
- these plots already show many interesting
conclusions (dfarrell07,
15:45:45)
- Luis is doing JVM monitoring in these jobs, so
can see CPU usage, have some graphs but bugs (unrelated to tests,
graphing/collecting/JVM stuff) (dfarrell07,
15:47:03)
- Some other person is working on making this JVM
stuff better, Luis reports (dfarrell07,
15:47:31)
- Will be pretty easy to change current
bulkomatic in-Java flow pushing to NB, just a single step in test
change (dfarrell07,
15:50:10)
- Will be able to do SB flow tests with CBench at
least, already have this, but there are better tools around and we'd
like to use them (dfarrell07,
15:50:37)
- Luis will go to tool owners (Veryx etc) and ask
for help/docs, see which are best (dfarrell07,
15:51:02)
- Discussion about using NSTAT in OF tests
(dfarrell07,
15:56:04)
- https://hub.docker.com/r/intracom/nstat/~/dockerfile/
link for NSTAT Dockerfile (dfarrell07,
15:56:08)
- https://github.com/intracom-telecom-sdn/nstat
NSTAT GitHub (dfarrell07,
15:56:19)
- Nikos of CPerf has been developing this tool,
has a new Dockerfile to make install easy, is very up for helping us
with it (dfarrell07,
15:56:41)
- Latte is another tool by Nikos and his
team (dfarrell07,
15:56:57)
- https://github.com/anastop/latte
Latte tool GitHub (dfarrell07,
15:57:14)
- Moving on to BGP/NetConf tests (dfarrell07,
15:58:43)
- mgkwill asks Vratko about which tests we want
to run, Vratko says there are some additional ones in CSIT but ones
in last whitepaper are most important (dfarrell07,
15:59:10)
- mgkwill has dev working with NetConf scale
stuff in internal lab, was hitting a bug, needed to add some auth
header, now running out of threads on test tool, some stuff about
OOM, it has 16GB, doing 8k or so, 10k was max Vratko et all could
get before, should maybe lower (dfarrell07,
16:01:25)
- mgkwill reports hitting those same issues with
Boron and Be SR3 (dfarrell07,
16:03:02)
- mgkwill and dfarrell07 talking about netvirt
scale tests (dfarrell07,
16:06:39)
- https://git.opendaylight.org/gerrit/#/c/26176/
(dfarrell07,
16:06:51)
- https://git.opendaylight.org/gerrit/#/c/29524/
Mininet, vs Docker networks scale stuff we can add netvirt to,
dfarrell07 will work on (dfarrell07,
16:12:57)
- mgkwill has been doing kolla scaling
(containers) for compute nodes, then running Rally and doing that
benchmarking, seeing varied results so no conclusions so far
(dfarrell07,
16:13:48)
- mgkwill is optimistic that we can get full
scale scale in next two weeks (dfarrell07,
16:14:31)
- Sajib asks about when running in cluster, basic
switch install is failing, asks if can bump heap size and stats of
VM, LuisGomez reports that yes you can select VM stats, Sanjib
reports it's not actually happening, Luis asks for runs in sandbox
to look at, will dig into (dfarrell07,
16:16:31)
- https://git.opendaylight.org/gerrit/#/c/37639/
Updated Mininet Patch - previous old version (mgkwill,
16:17:26)
- Sanjib asks about increasing heap size in
controller, Luis says that maybe should decrease number of flows as
well (at 100k now), also shows how to use a scriptplan to change
heap size (dfarrell07,
16:18:40)
- https://github.com/opendaylight/integration-test/tree/master/csit/scriptplans
Scriptplans, executed before tests (dfarrell07,
16:20:45)
- https://github.com/opendaylight/integration-test/blob/master/csit/scripts/controllermem_6g.sh
script to change mem (dfarrell07,
16:20:52)
- To change heap in test, make scriptplan that
points at that script (dfarrell07,
16:21:05)
- In a dramatic turn of events, Sanjib shows Luis
a major regression in OpenFlow Plugin, flow programming, Sanjib to
report bug ASAP (dfarrell07,
16:25:02)
- LuisGomez shows plotting docs (dfarrell07,
16:27:03)
- mgkwill asks about public cloud issues,
LuisGomez reports that issues seem to be much better last few days,
this was also thought on Int/* call yesterday (dfarrell07,
16:27:39)
- https://wiki.opendaylight.org/view/CrossProject:Integration_Group:System_Test:Step_by_Step_Guide#Optional_-_Plot_a_graph_from_your_job
Docs on how to make plots (dfarrell07,
16:28:15)
- https://hub.docker.com/r/vpickard/openvswitch/
container with OVS 2.3, 2.4 and 2.5, need Dockerfile (dfarrell07,
16:29:01)
- Want to move to this, but it's not really
worked into CSIT (dfarrell07,
16:29:16)
- Idea is to move from installing OVS in VMs to
these containers (dfarrell07,
16:29:36)
Meeting ended at 16:32:46 UTC
(full logs).
Action items
- (none)
People present (lines said)
- dfarrell07 (47)
- odl_meetbot (6)
- mgkwill (4)
Generated by MeetBot 0.1.4.