Vratko Polak

> in stats I guess exposing via get_stream_stats would be ok too, but as m_tx_queue_full is exposed via get_stats, m_tx_drop should be too.

Most values are identical, but few of them are different. Recent improvements made it less likely, but it is still possible for PDR to equal NDR. It took me a...

Still present [5] in rls2502. [5] https://s3-logs.fd.io/vex-yul-rot-jenkins-1/csit-vpp-perf-report-iterative-2502-3nb-spr/55/log.html.gz#s1-s1-s1-s3-s5-t1-k3-k7-k1-k1-k1-k8-k14-k1-k1-k1-k1

Still present [6] in rls2506. [6] https://logs.fd.io/vex-yul-rot-jenkins-1/csit-vpp-perf-report-iterative-2506-3n-alt/6/log.html.gz#s1-s1-s1-s3-s2-t2-k2-k10-k16

Still present [7] in rls2510. [7] https://logs.fd.io/vex-yul-rot-jenkins-1/csit-vpp-perf-report-iterative-2510-3n-alt/26/log.html.gz#s1-s1-s1-s3-s2-t2-k2-k10-k16

Marking FIB APIs as not thread safe avoids all the issues. I even managed to create a subtler fix [2] that seems to make the test pass, although it probably...

Welcome to github. In fd.io lab we now also have Emerald Rapids testbeds. 3n-emr is having some infra issues, butsing ad-hoc unidirectional ipsec suite (encryption only) I can confirm [8]...

[6] still shows this is present also in rls2502.

Present in rls2506, 3n-emr is now stable enough to confirm bad telemetry. [7] https://logs.fd.io/vex-yul-rot-jenkins-1/csit-vpp-perf-report-iterative-2506-3n-emr/13/log.html.gz#s1-s1-s1-s1-s16-t1-k3-k7-k1-k1-k1-k8-k10-k1-k1-k1-k14

I crated a Gerrit version [9] of the Intel fix. [9] 43707: crypto: call _mm256_zeroupper to fix SHA256 perf | https://gerrit.fd.io/r/c/vpp/+/43707