Sorry for late notice, discoveries were coming faster than I was able to investigate.
Not listing: - Regressions and failures with cause in CSIT code or drivers or firmware or hardware. - Regressions and failures identified but also already fixed on stable/2202. New in 22.02 (not present in 21.10):: Regressions: - Regression across all dpdk_plugin tests: - Cause is somewhere in DPDK upgrade, MTU rework or other related change. - In effect it disables the workaround for VPP-1876 tests are attempting to apply (setting high MTU). - The single VPP change where regression happens: https://gerrit.fd.io/r/c/vpp/+/34839 - More details: https://gerrit.fd.io/r/c/vpp/+/34839/4#message-d5d0c80dc32b1aa29a142fb48b6c132ca04ff332 - No easy fix in sight. - Regression in tests with l2-input graph node: - Cache miss introduced in https://gerrit.fd.io/r/c/vpp/+/34568 - More details: https://gerrit.fd.io/r/c/vpp/+/34568/2#message-63e1cdafce46e7d389ffb45d79df4fb622681888 - Revert ready: https://gerrit.fd.io/r/c/vpp/+/35164 - Due to misunderstanding, testing the revert begun late. - Some trouble verifying it restores L2 performance, maybe it introduces IP6 regression? - Regression in NAT44ED CSP tests: - Caused by a refactor/fix: https://gerrit.fd.io/r/c/vpp/+/34077 - Not much details beyond consistent performance difference. - No idea about fix. Failures: - Jumbo tests with higher encapsulation overhead: - Related to MTU, new default value is not high enough compared to what CSIT is attempting to set. - Also started by https://gerrit.fd.io/r/c/vpp/+/34839 - Also no easy fix in sight. Older issues not discovered before:: Failures: - Memif tests with jumbo frames: - Not clear yet if the cause is in VPP or in CSIT. - IMIX tests on 3n-icx fail for hmac512sha: - The 570 byte component of IMIX is dropped. - Does not happen on 3n-skx. - Does not happen for aes256gcm. - Also present with VPP 21.10 release code (we did no have Icelake testbeds when doing CSIT 2110 release). - Cause not identified yet. Other: - DPDK testpmd has performance spread higher than expected. - Sometimes there is multi-band structure visible. https://jira.fd.io/browse/CSIT-1798 - Affects also VPP vhost tests. - Latency spikes: - Not deterministic, frequency is log in general. - ARM and AMD seems to be affected more than Intel. Vratko.
-=-=-=-=-=-=-=-=-=-=-=- Links: You receive all messages sent to this group. View/Reply Online (#20888): https://lists.fd.io/g/vpp-dev/message/20888 Mute This Topic: https://lists.fd.io/mt/89319622/21656 Group Owner: vpp-dev+ow...@lists.fd.io Unsubscribe: https://lists.fd.io/g/vpp-dev/unsub [arch...@mail-archive.com] -=-=-=-=-=-=-=-=-=-=-=-