=================== #fdio-vpp: fdio-vpp =================== Meeting started by dmarion at 15:59:36 UTC. The full logs are available at http://ircbot.wl.linuxfoundation.org/meetings/fdio-vpp/2021/fdio_vpp/fdio-vpp-fdio_vpp.2021-02-09-15.59.log.html . Meeting summary --------------- * csit-2101 rls status: most benchmark testing for report completed, no major issues (mackonstan, 16:04:02) * draft version of csit-2101 report already online, incl. comparison tables to v20.09 (mackonstan, 16:04:16) * LINK: https://docs.fd.io/csit/master/report/ (mackonstan, 16:04:22) * LINK: https://docs.fd.io/csit/master/report/_static/vpp/performance-changes-2n-clx-xxv710-2t1c-pdr.txt (mackonstan, 16:04:30) * report publish target date 10-Feb, tomorrow, and we should be on time (mackonstan, 16:04:47) * some more updates going into rls notes and other static content e.g. methodology sections (mackonstan, 16:04:58) * After CSIT release is done, we will be moving to Ubuntu 20.04 as the main stack for benchmarking (mackonstan, 16:05:25) * There is also quite some work on the infra side, putting backend monitors in place, so that we can detect any issues early, thanks to Dave and Peter (mackonstan, 16:06:01) * Will come handy during the LFN/Vexxhost DC consolidation expected in end of Feb, Mar (mackonstan, 16:06:14) * svm area, custom queue implementation non-blocking when possible (DaveBarach, 16:08:15) * slight performance-at-scale improvement: 10% for NGINX tested w/ WRK (DaveBarach, 16:08:39) * next step: security improvements, private message queues for multiple applications (DaveBarach, 16:09:10) * DTLS patch, hard to test with iperf (DaveBarach, 16:10:01) * slightly less performant than tls. (DaveBarach, 16:10:47) * don't have support for retries w/ timeout (DaveBarach, 16:11:02) * SRTP trying to add support. Test difficulties adding RTP payloads (DaveBarach, 16:11:27) * no updates beyond fixing links on the main website (DaveBarach, 16:11:56) * Ray asks how we're planning to police doc / wiki bit-rot (DaveBarach, 16:13:54) * wiki is really hard to keep straight, should invest in the docset (DaveBarach, 16:14:29) * 21.01 shipped on schedule (DaveBarach, 16:15:02) * RC1 / RC2 timing worked out pretty well (DaveBarach, 16:15:21) * some last-minute churn, probably need to be more strict about avoiding churn post-RC2 for 21.06 (DaveBarach, 16:16:00) * data center move; unless someone insists on point releases, let's wait (DaveBarach, 16:16:49) * folks interested in Centos7 need to figure out where to go since Red Hat is moving away from Centos (DaveBarach, 16:18:41) * DaveW: Centos7 is too old to add to our CI (DaveBarach, 16:19:41) * Damjan: really need a list of package dependencies, vs. distro versions (DaveBarach, 16:20:58) * same issue occurred with Debian9 vs Debian-10 (DaveBarach, 16:21:53) * ACTION: DaveW and Andrew to investigate making minimum version listing (DaveBarach, 16:22:29) * discuss again next time (DaveBarach, 16:22:43) * infra breakages over the last month, dealt with, ping DaveW directly with ongoing infra issues (DaveBarach, 16:23:39) * auto-abandon proposal / RFC. We have 600+ patches in the gerrit queue (DaveBarach, 16:24:09) * costs cycles to have the queue that large (DaveBarach, 16:24:24) * auto-abandon would address that aspect of the problem (DaveBarach, 16:24:47) * Seems like most folks happy to enable auto-abandon, except for Paul V, Ole (DaveBarach, 16:26:19) * of the outstanding patches, 477 are waiting for author actions (address comments, rebase, etc.) (DaveBarach, 16:26:52) * 42 awaiting review (DaveBarach, 16:26:58) * doing a reasonable review job (DaveBarach, 16:27:53) * ACTION: send out list of patches to clean up (DaveBarach, 16:36:48) * ACTION: Dave W to see if converting patches to work-in-progress will fix the problem (DaveBarach, 16:39:13) * interesting exercise to run tests w/ 2 workers (DaveBarach, 16:40:36) * patch to run this way by environment variable pushed (DaveBarach, 16:41:04) * another patch in the works which tags tests which don't work multicore (DaveBarach, 16:41:58) * one reason: tests which assume only one set of counters, blow up due to split counts (DaveBarach, 16:42:25) * shout-out to Dave W for CI failure cause enhancements (DaveBarach, 16:46:00) * aarch64 scalable vector enhancements (DaveBarach, 16:46:59) * discuss changes in vpp to support / use these instruction-set changes (DaveBarach, 16:47:25) * instrs are vector-length agnostic (DaveBarach, 16:49:53) * coding style involves looping (DaveBarach, 16:53:36) * topic really needs a full hour (DaveBarach, 16:55:55) * ACTION: meet tomorrow same time slot to continue (DaveBarach, 16:58:18) Meeting ended at 16:58:46 UTC. People present (lines said) --------------------------- * DaveBarach (49) * mackonstan (9) * collab-meetbot (4) * dmarion (2) Generated by `MeetBot`_ 0.1.4