#fdio-csit: FD,io-CSIT project meeting
Meeting started by mackonstan at 15:03:07 UTC
(full logs).
Meeting summary
-
- Vratko Polak. (vrpolak,
15:03:15)
- Dave Wallace (dwallacelf,
15:03:19)
- Tibor Frank (tifrank,
15:03:19)
- Jan Gelety (jgelety,
15:03:21)
- Agenda bashing (mackonstan, 15:04:50)
- Physical labs, testbeds (mackonstan, 15:06:22)
- no issues (mackonstan,
15:06:25)
- Inputs from projects - VPP (mackonstan, 15:06:46)
- Maciek - MLX native code improvements are now
merged - we should see this in trending tests (mackonstan,
15:07:29)
- DaveW - lb perf job not running correctly in
20.01. VPP issue identified, to be cherrypicked, and then perf
tested. (mackonstan,
15:11:37)
- Inputs from projects - TSC (mackonstan, 15:11:50)
- discussion about FD.io dev event collocated
with DPDK dev event in Bordeaux France in October (mackonstan,
15:12:55)
- re scapy/gpl2 - awaiting status from Ray
(mackonstan,
15:13:54)
- Inputs from projects: CNCF CNF Testbed (mackonstan, 15:14:24)
- Maciek has an action to check status with
Michael Pedersen re status (mackonstan,
15:14:48)
- Inputs from projects: VSAP (mackonstan, 15:16:01)
- request for adding tests for VSAP - email from
Yulong (mackonstan,
15:17:18)
- VSAP code in repo (mackonstan,
15:18:09)
- https://git.fd.io/vsap/tree/
(mackonstan,
15:18:28)
- VSAP mailer missing, project meetings
missing (mackonstan,
15:19:20)
- Releases (mackonstan, 15:19:36)
- Releases CSIT-2001 (mackonstan, 15:20:02)
- Jan - 2n-skx testing: 9/10 repeatibility tests
are done, 1/10 in progress ETA shortly/today; nfv-density 1/10
started; full-tests to be scheduled. (mackonstan,
15:22:35)
- Jan - 3n-skx testing: 5/10 repeatibility tests
are done; full-tests to be scheduled. (mackonstan,
15:23:30)
- Analysing results - Tibor - all available data
is in draft report in master (mackonstan,
15:24:04)
- https://docs.fd.io/csit/master/report/
(mackonstan,
15:24:21)
- quite a lot of performance changes vs. previous
release (mackonstan,
15:30:07)
- avf ip4base regressions per known issue with
Xeons/FVL NICs (mackonstan,
15:30:47)
- nat44 regression (mackonstan,
15:30:57)
- and generic issue with NDR rates being bad, due
to Xeon Skx/FVL NICs environment that's under investigation
Intel\ (mackonstan,
15:32:02)
- ipsec numbers lower - need to
double-click (mackonstan,
15:33:37)
- hoststack tests - DaveW: Jan reporting latest
test code failing, DaveW couldn't find the root cause as locally
perf tests works, related to interface link initialization, 3n-hsw
xl710 dpdk driver. Jan+DaveW collaborating to fix this. (mackonstan,
15:36:52)
- targeting maintenance report release tomorrow -
with skx results. (mackonstan,
15:38:38)
- next week maintenance with full set of tests on
skx, fixed hoststack, fixed lb (mackonstan,
15:38:52)
- Releases - CSIT-2005 (mackonstan, 15:39:23)
- https://wiki.fd.io/view/CSIT/csit2005_plan
(mackonstan,
15:39:34)
- CSIT in AWS and Azure code is pushed in
(mackonstan,
15:40:02)
- Operations (mackonstan, 15:40:28)
- DaveW - after naginator removal, things work -
no complaints, no increase of rechecks noticed (mackonstan,
15:42:29)
- Vratko - vpp-csit-device tests - re-enable
voting - gated upgrade to api-crc check process required
(mackonstan,
15:43:18)
- Vratko will provide a link to the change of
api-crc process (mackonstan,
15:44:32)
- Centos CI - not running in vpp-device - stale
now (mackonstan,
15:44:42)
- https://gerrit.fd.io/r/c/csit/+/22354
<-- Plan for more advanced API change process. A prerequisite for
vpp-csit devicetest voting. (vrpolak,
15:45:15)
- CSIT committers - candidates: Juraj, Yulong.
Maciek to check gerrit/git stats and start voting email on csit-dev
list (mackonstan,
15:47:37)
- VPP code performance (mackonstan, 15:47:53)
- Peter - vhost vm tests failing - after upgrade
of default dpdk 20.02, mergeable buffer, being fixed (mackonstan,
15:49:10)
- - avf - interface init issues - 3 separate
intermittent issues. Only affecting skx, no clx. (mackonstan,
15:51:11)
- Peter - rdma tests failing - due to typo in qsz
change (mackonstan,
15:51:41)
- Vratko - identify two anomalies: 1. ipsec hw -
after fix difference in performance in 100s% for some tests, culprit
patch identified; 2. ipsec sw - 10-12% regression for two out of
four algos, , culprit patch identified (mackonstan,
15:53:54)
- Developments (mackonstan, 15:56:49)
- Balaji - py utility generating gcov data - easy
to parse output. Will announce to vpp-dev, csit-dev. (mackonstan,
15:58:45)
Meeting ended at 15:59:19 UTC
(full logs).
Action items
- (none)
People present (lines said)
- mackonstan (49)
- collabot` (4)
- vrpolak (2)
- dwallacelf (1)
- jgelety (1)
- tifrank (1)
Generated by MeetBot 0.1.4.