-
Task
-
Resolution: Unresolved
-
Major
-
None
-
None
-
5
-
False
-
-
False
-
-
None
-
rhel-net-ovs-dpdk
-
-
This ticket is created automatically to track an investigation spike related to the issue: FDP-1232.
The goal is to identify the root cause or clarify unknowns and evaluate possible solutions or workarounds for the problem described below.
Description of problem:
When measuring VM-to-VM TCP Stream throughput, localnet attachment types can achieve expected throughput, however localnet specifically shows significantly higher retransmit values compared to other net types. Note that this behavior is not specific to localnet on br-ex, it is observed when attached to another OvS bridge as well.
Version-Release number of selected component (if applicable):
4.18.1
How reproducible:
Multiple runs on compact cluster, will confirm on full worker cluster
Steps to Reproduce:
Full details in this perf report:
https://docs.google.com/document/d/1pDOjROxCqrfNhE2nAkeJaRen3RtfixeJdt82ae_aW20/edit?usp=sharing
Actual results:
TCP retransmit values should be somewhat similar to other net types (up to ~40K for test case)
Expected results:
TCP retransmit values for localnet are ~100K-400K for test case
Additional info:
Affected Platforms: baremetal Contact jhopper@redhat.com for direct cluster access, I will continue to debug and post updates here otherwise. * traffic path: vm to vm