====================================== #fdio-csit: FD.io CSIT project meeting ====================================== Meeting started by mackonstan at 15:00:19 UTC. The full logs are available at http://ircbot.wl.linuxfoundation.org/meetings/fdio-csit/2020/fd_io_csit_project_meeting/fdio-csit-fd_io_csit_project_meeting.2020-11-18-15.00.log.html . Meeting summary --------------- * Dave Wallace (dwallacelf, 15:00:34) * Juraj Linkeš (jlinkes, 15:00:34) * Tibor Frank (tifrank, 15:00:41) * Vratko Polak. (vrpolak, 15:00:46) * Jan Gelety (jgelety, 15:01:14) * Agenda bashing (mackonstan, 15:02:37) * Juraj, want to talk about perf tests on Arm/Tishan, and how to make them execute faster. (mackonstan, 15:03:34) * Inputs from projects - VPP (mackonstan, 15:04:03) * Andrew, reminder sent F0 in 3 weeks (mackonstan, 15:04:34) * Maciek, questions from community about vpp memory/heap/hugepages configuration, asking for clearer usage guidelines. Looking for volunteers to work with me on that. (mackonstan, 15:07:50) * Inputs from projects - TSC (mackonstan, 15:08:09) * Vratko, CSIT master branch is using GPL license in GPL directory. (mackonstan, 15:08:29) * Vratko, VPP master is still in works, RayK holds the pen to complete drawing the plan to conclude the Scapy driven part of make test GPL re-license. (mackonstan, 15:09:55) * Releases CSIT-2101 (mackonstan, 15:10:28) * F0 milestone 2020-12-09 (mackonstan, 15:11:22) * Vratko, 2101 is the first rls with new VPP API deprecation process kicks in (mackonstan, 15:12:03) * Vratko, one safety net is voting vpp-csit-device job in vpp (mackonstan, 15:13:23) * Andrew, here is the pointer to **all** APIs that will get removed in v21.01, https://docs.fd.io/vpp/20.09/de/d1e/release_notes_2009.html > Newly deprecated API messages (mackonstan, 15:16:05) * Maciek, Jan, obvious CSIT test exposure: ipsec_tunnel_if_add_del (mackonstan, 15:17:11) * Maciek, Jan, Vratko to skin this cat. (mackonstan, 15:17:48) * Releases CSIT-2009 (mackonstan, 15:18:32) * Tibor, no new data/content today, so no report update (mackonstan, 15:19:05) * Tibor, nat44 test results and methodology, are ready to be pushed into rls2009 report. Data is already published in the master version. (mackonstan, 15:20:15) * Tibor, will wait till tomorrow evening to see if the gerrit methodology secions' review is complete, and if it is, publish the report (mackonstan, 15:21:47) * Tibor, started to test 2n-zn2 … (mackonstan, 15:24:43) * Peter, BUT, there is an issue with uploading artefacts to Nexus (mackonstan, 15:25:05) * Dave, discussed with Vanessa, unclear where the problem lies (mackonstan, 15:26:32) * Dave, jenkins CI is now down for all vpp and csit jobs for 24hrs (mackonstan, 15:30:58) * Dave, will bring it up on TSC call tomorrow (mackonstan, 15:31:27) * Vratko, sandbox Jenkins works though (mackonstan, 15:31:47) * Dave, will ask for a full visibility into Jenkins configs (on TSC call tomorrow), as currently blind (mackonstan, 15:35:25) * Physical Infrastructure - Xeon, Atom (mackonstan, 15:35:44) * Maciek, Peter, will discuss offline best way to install 6x E810 NICs in 2n-clx testbeds (mackonstan, 15:36:38) * Physical Infrastructure - Arm testbeds (mackonstan, 15:37:00) * Juraj, why do perf tests twice as long to execute on Arm/Tishan vs Xeon, and how to make them execute faster. (mackonstan, 15:37:45) * Juraj, one possible reason is that they are running on Arm executors that are slower (mackonstan, 15:38:14) * Juraj, but switching to ThunderX2s, fast servers, the jobs still take longer (mackonstan, 15:38:42) * Juraj, but turning off SMT on ThunderX2 executors, brought the execution time in line with Xeons (mackonstan, 15:39:18) * Juraj, root caused to slower cores on ThunderX1s. ThunderX2s are fine. (mackonstan, 15:40:09) * Juraj, two options to address this going forward: i) use Xeon executors, would need to add Arm arch support; ii) use ThunderX2 servers as executors, but there are only two servers. (mackonstan, 15:41:27) * Vratko, Peter, clearly prefer option ii) due to its simplicity, e.g. avoids cross-compiling (mackonstan, 15:43:56) * Juraj, email csit-dev to drive consensus (mackonstan, 15:46:29) * Physical Infrastructure - AMD (mackonstan, 15:47:42) * Maciek, looking good. (mackonstan, 15:49:02) * Maciek, re FD.io lab extension in 2021, would like to propose to set up TWS starting next week (mackonstan, 15:55:30) * Maciek, expecting data points re existing costs for next week (mackonstan, 15:55:51) * Virtual Infrastructure, CI/CD (mackonstan, 15:56:05) * Juraj, arm-based vpp-csit-device jobs are in and working, but not voting (mackonstan, 15:56:29) * Maciek, let's wait for a good green period before enabling voting (mackonstan, 15:57:54) * Agreed to add this as a standing point to agenda (mackonstan, 15:58:51) * Environment, Framework (mackonstan, 15:59:02) * Maciek, ip6 latency tests broken, TRex ver at fault (mackonstan, 15:59:16) * Vratko, Peter, TRex 2.86 works, so moving to that (mackonstan, 15:59:45) * Performance Tests (mackonstan, 16:01:57) * Vratko, nothing new since last week (mackonstan, 16:02:04) Meeting ended at 16:02:24 UTC. People present (lines said) --------------------------- * mackonstan (58) * collab-meetbot (14) * dwallacelf (1) * jlinkes (1) * tifrank (1) * vrpolak (1) * jgelety (1) * pmikus (0) Generated by `MeetBot`_ 0.1.4