-
Bug
-
Resolution: Can't Do
-
Major
-
None
-
None
-
Quality / Stability / Reliability
-
0.42
-
False
-
-
False
-
None
-
This issue is security relevant
-
-
Low
-
None
Description of problem:
IO tests with CNV VM shows bigger netowrk latency than tests executed inside pods. OpenShift version: 4.16.9 OpenShift Virtualization 4.16.4 We ran fio inside vm in first case and inside pod for second case, below are test results vm case : iops: 110 ( bs 4k , but similar situation for other block sizes ... ) latency: 9083463 ( us ) ~ 9s pod case : iops: 758 ( bs 4k ) latency: 1316564 ( us ) ~ 1.3s fio job: randwrite, filsize=10GB, ioengine=libaio, bs=4k, iodepth=1, numjobs=10, time_based=1 Version-Release number of selected component (if applicable):
OpenShift version: 4.16.9 and OpenShift Virtualization 4.16.4
How reproducible:
Always in our testing.
Steps to Reproduce:
Get CNV environment, test IO inside VM and inside pod when both VMs and Pods are using same storage class - it should not matter what storage class is in my opinion. Test at least 10 fio clients. Compare results. We ( perf / scale ) team have working cluster in place which can be used for bz verification. We are also ready and will help with testing. Please let us know.
Actual results:
Expected results:
Additional info: