neutron

  1. Bug #1918145
  2. Comment #34

Comment 34 for bug 1918145

Revision history for this message
Tyler Stachecki (tstachecki) wrote :

Some more performance data:

Before the patch, on an internal test cluster, we observed that the database would go underwater (~0% host CPU idle) when creating/deleting even 100 VMs.

With the patch, I was able to create/delete 1k+ VMs faster than I could 100 VMs before the patch. Following that, the database load dropped from 4000% CPU (pre-patch) to only 1000% after... while doing 10x the amount of work (1k VMs vs. 100).

The creation/deletion of 1k VMs probably could have gone faster yet, but the bottleneck shifted somewhere else in the control plane; the database is now no longer a limiter.