========================================= #fdio-meeting: FD.io CSIT Project Meeting ========================================= Meeting started by mackonstan at 15:00:00 UTC. The full logs are available at http://ircbot.wl.linuxfoundation.org/meetings/fdio-meeting/2022/fd_io_csit_project_meeting/fdio-meeting-fd_io_csit_project_meeting.2022-02-23-15.00.log.html . Meeting summary --------------- * Maciek Konstantynowicz (mackonstan, 15:00:10) * Tibor Frank (tifrank, 15:01:58) * Agenda bashing (mackonstan, 15:02:00) * Vratko Polak. (vrpolak, 15:02:06) * Inputs from projects - VPP (mackonstan, 15:03:30) * Fan Zhang (fan_zhang, 15:03:41) * Andrew, v22.02 artifacts being pushed to packagecloud (mackonstan, 15:03:59) * Andrew, into rls directory, should be done in couple hours (mackonstan, 15:04:16) * Andrew, few fixes came late, yday, i) vxlan fixing a crash in masterthat changed meaning of api and introduced another problem, now reverted; ii) mtu callback related for memif, virtio, may have meaningful impact on jumbo tests; (mackonstan, 15:09:06) * Fan, related to above, chatted with dpdk dev who authored mtu/mfs changes in dpdk. this was done to unify mfs/mtu handing across multiple nic models and taking care of encap/decap offload in nic's (mackonstan, 15:10:37) * Dave Wallace (dwallacelf, 15:12:08) * Fan, re dpdk performance degradation report by Vratko in vpp-dev: this could, but because of defaulting to 1518B mfs/mtu, avx2 rx_burst functional call in linear mode (using couple of mbufs, with additional check), previously vpp was receiving 256 frames in one go, now it's 64 frames. (mackonstan, 15:15:26) * Peter, looking at dpdk init debug, after interfaces go up, one can see dpdk rx_burst working with avx2 in linear mode (mackonstan, 15:17:24) * Fan, need to double click if it's really the root cause behind the vpp rc2 regressions reported by Vratko on vpp-dev, 5-15% (feature-simple tests) throughput drop (mackonstan, 15:19:22) * Vratko, Maciek, I/O "heavy" regression, the more features one uses, the lower the regerssion; the simpler the config the higher regression. (mackonstan, 15:20:04) * Inputs from projects - TSC (mackonstan, 15:20:50) * Maciek, Dave, hw budget allocation for dram for arm build servers, and mlx connectx5s for trex 100ge nics for clx testbeds (mackonstan, 15:23:02) * Dave, TSC chair nomination period is coming to an end this week (mackonstan, 15:23:43) * Dave, this will be followed by voting among TSC members (mackonstan, 15:23:56) * New or (to be) refactored tests (mackonstan, 15:24:20) * VPP RC1, RC2 hot patch(es) items (mackonstan, 15:25:28) * Peter, i) jumbo frames are supported in af_xdp kernel code up to "memory alignment" ~3kB, no support for 9000B; ii) jumbo frames issue with vpp memif, bug to be tracked re support mtu; (mackonstan, 15:28:13) * Vratko, iii) found the place where vpp sets default IP MTU to 9000B. (mackonstan, 15:29:46) * Maciek, suggesting to discuss on csit-dev mailer how to re-design existing csit jumbo frame size tests (mackonstan, 15:31:11) * Releases CSIT-2202 (mackonstan, 15:31:35) * Tibor, all planned rc2 iterative tests are done; all but 2 coverage tests are done too. (mackonstan, 15:32:01) * Tibor, all data is in master ver of report (mackonstan, 15:32:12) * Tibor, failed tests, nothing unexpected, only known issues. Vratko working thru them, Viliam composed summary, sent to csit-report. (mackonstan, 15:34:21) * Tibor, draft report dynamic content validation, still to be completed fully (mackonstan, 15:35:31) * Tibor, 2n-cix and 3n-icx testbeds added (mackonstan, 15:35:50) * Tibor, static content updates - rls notes, RCA (mackonstan, 15:36:08) * Tibor, static content - need to update environment versions (mackonstan, 15:36:26) * Tibor, 1st rls results expected to be making its way into master report draft on Mon or Tue next week (mackonstan, 15:37:42) * Vratko, commenting on rc2 regressions, per email to vpp-dev (mackonstan, 15:38:28) * LINK: https://lists.fd.io/g/vpp-dev/message/20888 (mackonstan, 15:38:34) * Vratko, tsh tests are affected by TRex calibration … (mackonstan, 15:38:59) * Vratko, for some regressions need more then 5 iterative runs (mackonstan, 15:41:08) * Physical Infrastructure (mackonstan, 15:47:13) * re delayed shipment of supermicro systems for ICX - 2 units expected next week per vexxhost ticket (mackonstan, 15:49:56) * LINK: https://support.vexxhost.com/hc/en-us/requests/361959 (mackonstan, 15:50:03) * Peter, 2 ampere perf servers are in transit per latest DHL updates (mackonstan, 15:50:52) * Virtual Infrastructure, CI/CD (mackonstan, 15:51:39) * Dave, Peter, s3 migration is happening in the background, vpp docs are taking time, csit is done (mackonstan, 15:52:34) * Environment, Framework (mackonstan, 15:52:47) * Andrew, Vratko, been discussing improvements in the are of CSIT CLI and API usage, and tracking VPP API changes in automated fashion (mackonstan, 15:54:29) * Trending (mackonstan, 15:55:26) Meeting ended at 16:12:32 UTC. People present (lines said) --------------------------- * mackonstan (45) * collab-meetbot (4) * tifrank (1) * vrpolak (1) * fan_zhang (1) * dwallacelf (1) Generated by `MeetBot`_ 0.1.4