#opnfv-cperf: cperf
Meeting started by dfarrell07 at 18:17:05 UTC
(full logs).
Meeting summary
-
- Welcome back jamoluhrsen, mgkwill :)
(dfarrell07,
18:17:30)
- dfarrell07 and Jamo talk about Hello World job
triggering (dfarrell07,
18:25:19)
- dfarrell07 thought Apex should be triggering
it, but that's not true, it's manual (dfarrell07,
18:25:36)
- but jamoluhrsen checked manual triggering and
that fialed, pod 2 is dead (dfarrell07,
18:25:51)
- mgkwill notes that pod was online for a while,
then back off -what's up? (dfarrell07,
18:26:15)
- jamoluhrsen will work on getting job running
against a different pod, or working pod 2, next week (dfarrell07,
18:26:36)
- dfarrell07 gives overview of Nikos recent
switch scale results (dfarrell07,
18:29:23)
- Vary switches, number of containers, types of
isolation (vms vs containers) (dfarrell07,
18:29:46)
- VMs and containers scale about the same, which
is good for us in OPNFV (dfarrell07,
18:29:58)
- opnfv test tools typically go in containers, so
that works for us (dfarrell07,
18:30:19)
- more isolation gives better results
(dfarrell07,
18:30:28)
- future work around keeping container count
high, and changing the rate of switch add and group size to find
quickest way to scale to largest number (dfarrell07,
18:31:01)
- LuisGomez talks about infra setup, how we might
could do this in ODL (dfarrell07,
18:31:24)
- LuisGomez talks about how to very number of
switches in these scale test, Nikos answers that they count
switches in ODL datastore (dfarrell07,
18:33:18)
- none of these tests add hosts, LuisGomez talks
about different tests that add hosts (dfarrell07,
18:33:53)
- LuisGomez reports that host tests would require
L2Switch, which isn't so good, so they don't use in CSIT (maybe
demos) (dfarrell07,
18:34:18)
- Nikos talks about how they currently vary many,
many params when doing NSTAT tests, basically all combinations.
Thinking about moving to smarter techniques sub-sampling tech, like
hill climbing to find local maxes (dfarrell07,
18:35:43)
- this better to find limits (dfarrell07,
18:36:09)
- august is vacation month in much of EU, Nikos
reminds ;) (dfarrell07,
18:37:39)
- Nikos resumed work on latte this week, has made
good progress, has logic to match pkt ins with flow mods, written
unit tests that work, seeing some errors when running IT with real
multinet traffic (dfarrell07,
18:38:33)
- Nikos is also working on decoupling main logic
and logic that extracts matching fields from matches, user could
write those extraction logic things and tool could use them to look
at arrival times, latencies, etc (dfarrell07,
18:39:25)
- al and nikos talk about which fields to use, as
transaction ID, buffer ID, etc would not work and couldn't be made
to work by changing ODL, Nikos is using src and dst macs in pktins
and flowmods (dfarrell07,
18:40:07)
- Nikos also using IP and port number of
switches, to be more sure (dfarrell07,
18:40:20)
- dfarrell07 asks mgkwill about status of netvirt
scale testing intel is doing (dfarrell07,
18:42:34)
- Matt W on call has been working on this
(dfarrell07,
18:42:48)
- have scaled up to 40+ compute nodes on one
setup, mgkwill says (dfarrell07,
18:43:09)
- matt gets mic, has been bringing up VMs with
virst/libvirt, use those as hosts, spin up instances in those, have
been looking at using containers as alt (dfarrell07,
18:43:35)
- matt is spinning up legit compute nodes with
all openstack services, running nova compute nodes in VMs
(dfarrell07,
18:44:13)
- Another person has 3 physical systems with 20+
compute nodes each (dfarrell07,
18:44:30)
- mgkwill talks about moving all this to use
containers (dfarrell07,
18:44:37)
- spin up as many compute nodes as possible, hook
up to odl, see what blows up (dfarrell07,
18:44:57)
- mgkwill has golden number of 500 or so
(dfarrell07,
18:45:03)
- would then spin up networks, how long that
takes with all these nodes already up, etc (dfarrell07,
18:45:27)
- first step is to get to that scale, 500+
(dfarrell07,
18:45:37)
- mgkwill has been working on spinning up ODL
with Rally, would give some more info (dfarrell07,
18:46:41)
- dfarrell07 is now catching up on notes from
start of meeting, when he was talking not writing (dfarrell07,
18:48:12)
- dfarrell07 has been talking to David, OPNFV Rel
Manager, about CPerf in/out of C-release (dfarrell07,
18:48:20)
- David was interested in how ODL releases work
with testing projects (dfarrell07,
18:48:30)
- dfarrell07 described ODL SRs, how we have
SRs/milestones for parts of ODL that are tightly coupled, have API
agreements they need to make (dfarrell07,
18:48:34)
- ODL's testing, packaging and similar projects
don't take part in SR, because they need to work with all
releases/branches (dfarrell07,
18:48:41)
- ODL test projects work with master branch only,
and change test code to run against various releases if/where
necessary (dfarrell07,
18:48:46)
- David says that OPNFV TSC has been looking at
different sets of milestones for test projects, or generally making
the release process work better for test projects (dfarrell07,
18:49:02)
- dfarrell07 ended up talking about this on
TestPerf subgroup call this morning (dfarrell07,
18:49:07)
- Al, David, dfarrell07 talked about using JIRA
to track progress, not having milestones/SR overhead (dfarrell07,
18:49:12)
- dfarrell07 wrote updates for all JIRAs with a
quick dump of progress. Please go add details/info to ones you own,
and let's all try to keep them up-to-date. (dfarrell07,
18:49:19)
- Keeping JIRAs up-to-date is our main way to
communicate with OPNFV community about what we're up to (dfarrell07,
18:49:26)
- Ran out of time, here are some teasers for
stuff we didn't have time to talk about (dfarrell07,
18:50:23)
- Great discussion on integration-dev mailing
list about Luis' OpenFlow tests, feedback from Al et al about
timing/metrics/tests (dfarrell07,
18:50:25)
- https://wiki.opnfv.org/display/yardstick/How+to+store+other+project%27s+test+results+in+InfluxDB
New wiki about how to store results in OPNFV DB (dfarrell07,
18:50:30)
- https://etherpad.opnfv.org/p/qtip-overview
QTip asks to give feedback about our projects and how they
can/should interact with QTip (dfarrell07,
18:50:34)
- dfarrell07 (well, CPerf) is scheduled to chair
TestPerf meeting next week. Yay. :| (dfarrell07,
18:50:39)
Meeting ended at 18:50:45 UTC
(full logs).
Action items
- (none)
People present (lines said)
- dfarrell07 (53)
- collabot (3)
- mgkwill (1)
Generated by MeetBot 0.1.4.