> 3n-hsw
Around 2nd of April there happened a regression
in eth-l2bdscale1mmaclrn test, visible here [0].
Trending needed multiple runs to identify it is there,
and alerting is configured not to report "old" regressions
(so it did not report this one).
Anyway, the closer look shows previously the
> 100ge2p1cx556a-rdma
No VPP code change.
Up to now, RDMA tests were suffering from "duration stretching".
That is a phenomenon where TRex takes longer to send
the required number of packets, so device under tests
experiences smaller offered load and MRR value is inflated.
We have suppressed the p
