Skip to content

3n-icx TB38 e810 has poor PDR results with jumbo #4167

Description

@vrpolakatcisco

This affects coverage test performance for rls2606. Testing on master confirmed TB37 has good performance [0] (near 120 Gbps) but TB38 had bad performance [1] (under 50 Gbps). The coverage run shows all ip4 jumbo tests are affected (including scale tests and both drivers) but stats trials usually see zero loss. But I have found one with considerable loss [2], but telemetry [3] implies packets are lost before of between VPP instances (thus not a duration stretching issue) without any visible RX or TX error and low Vectors/Call values (so CPU does not seem to be a bottleneck).

[0] https://logs.fd.io/vex-yul-rot-jenkins-1/csit-vpp-perf-verify-3n-icx/30259791753/log.html.gz#s1-s1-s1-s1-s1-t1
[1] https://logs.fd.io/vex-yul-rot-jenkins-1/csit-vpp-perf-verify-3n-icx/30259803768/log.html.gz#s1-s1-s1-s1-s1-t1
[2] https://logs.fd.io/vex-yul-rot-jenkins-1/csit-vpp-perf-report-coverage-2606-3n-icx/28566594637/log.html.gz#s1-s1-s1-s1-s10-t6-k2-k10-k20-k13-k1-k2
[3] https://logs.fd.io/vex-yul-rot-jenkins-1/csit-vpp-perf-report-coverage-2606-3n-icx/28566594637/log.html.gz#s1-s1-s1-s1-s10-t6-k2-k10-k20-k14-k1-k1-k1-k1

Metadata

Metadata

Assignees

No one assigned

    Labels

    No labels
    No labels

    Type

    No type

    Projects

    No projects

    Milestone

    No milestone

    Relationships

    None yet

    Development

    No branches or pull requests

    Issue actions