Hi All,
I was patching our ESXi cluster (10 hosts) this week (patching 2 hosts at a time) and I have noticed two instances of linux VMs (one RHEL 5, one RHEL 6) running very slowly after being located on an ESXi that had high load (CPU/Memory in the high 90's).
After relocating the VM off the host (via DRS or manually), the VM still runs very slowly. There is plenty of idle compute capacity (CPU/RAM) on the hosts, and looking at the performance via vCenter shows a VM with ~25% load. Looking within the VM (via munin or TOP) shows extremely high CPU load (~200). I don't see any CPU ready time from outside the VM.
My thoughts are when the VM was on the host with CPU contention, the VM may have slowed down (or throttled the guest vCPU). The issue is the guest has not recovered from this (or reverted back to full performance). My thoughts are along the lines of a laptop with speedstep/power management, that throttles the CPU when running on battery, but runs 100% on AC power.
Rebooting the guest OS returns the VM to full performance.
How can i troubleshoot this/gather evidence/prevent this from happening? with over 600 VMs on this cluster (a blend of Windows and Linux, approx 200 Linux/400 Windows) only 2 VMs have exibited this behaviour.
Regards,
Glynn
