Degraded performance when creating new VMs

[vk-en]

The problem is that if you create a VM, run a load on it (for example, a performance test), wait for the performance test to complete, recording the results. Delete the created VM. Repeat all the same again, and for example, by the 4th iteration, we can get a performance degradation of up to 70%, with the same test, in comparison with the first created VM. This behavior has been seen on GCP. I have not tested it on other platforms.

[giggsoff]

Please elaborate, what is "performance degradation" in your case? Performance of Eden side (test or perhaps building time) or performance of EVE on gcp (which may be affected by unpredictable things inside their hypervisor)?

[vk-en]

Degradation of I/O performance (in my case, as part of an io_performance test). The observation was done at GCP

[rvs]

Two suggestion @vk-en

1. try to see if you can reproduce it on bare metal EVE deployment (non-GCP)
2. lets try to figure out why this may be the case to begin with

[giggsoff]

On hardware: first run: https://openbenchmarking.org/result/2102011-HA-TESTEVEFI46 second run: https://openbenchmarking.org/result/2102015-HA-TESTEVESE17 the third: https://openbenchmarking.org/result/2102013-HA-TESTEVETH38

[rvs]

This is really bizarre -- are you guys digging into this?