So I did some tests a while back in my test lab, using CPU-bound test VMs to see how the different levels of fully-automated DRS would work... And I was generally pleased with the results... Conservative really was conservative... so I have had no problems making DRS fully automatic throughout my VMware environment...
Well, today, we had a production host server fail (memory error). I'm happy to report that HA worked flawlessly... All 20 or so VMs failed over to other hosts in less than a minute....
So, we fixed the host server and put in back in the cluster... and we were all waiting for the DRS magic to begin... but not a single VM was moved...
This particular cluster had 4 host servers... DL585 G1s 4-proc (dual-core). 90 VMs. So about 30 each on 3 hosts. CPU was sitting at 18%, 25%, and 12% for those three.. but memory utilization was at 90%, 86%, and 85%... so we could definitely use some load-balancing onto the fourth server that was added back into the cluster.
DRS was set at conservative by the way... 4 stars or better to move...
Ten minutes later... not a single VM had been moved... We moved one manually to make VMotion was working properly...
We set DRS to 3 stars or better to move... Still not a single VM moved...
Set DRS to 2 stars or better to move... Finally it moved 5 VMs...
So now we're sitting at
Host 1 17% CPU 89% memory
Host 2 12% CPU 85% memory
Host 3 11% CPU 73% memory
Host 4 2% CPU 10% memory
Sure seems that at 2 stars (moderate improvement to move), we'd see more movement...
Can anyone explain better how this works? What's the best way to get the 4 host servers back to roughly equal loads again? Other than manually moving another 15-20 VMs...
