An unbelievable time was had by our two "teams" at the VMware training. We did our best to cripple the VI kernel and eventually succeeded. In all actuality, the process proved what a terrific product VI3.5 really is.
If you can imagine 31 VMs running processor stress tools then you can start to get a flavor of what we were up to. Picture two DL360’s (presumable G4’s) running dual 3.7 procs. Then picture 2GB of ram each (blah) and 31 VMs running in a Vmotion/HA configuration. All the while picture the VMs running processor stressing tools. Then picture VMware’s HA and Vmotion constantly recalculating the right setup for the cluster.
We had this going on for 36 hours straight averaging 97% processor utilization and 90% ram utilization. We didn’t bother measuring the disk latency which I’m sure was unacceptable, but that didn’t matter. VMware performed flawlessly. At one point, we tried to PSOD (Purple screen of death) one of the hosts. We couldn’t do it. However, we were able to choke the system(s) into submission. One of our team (Dana) implemented a known linux command to chew up all the ram remaining on the box. It just put the system into a "hung" state. Vmotion nor HA was working on that particular host at that point. When we rebooted the host, the VMWare and HA started moving the VMs until the server was back online.
Very cool test that some may not have ever tried. It’s also something to consider when designing a failover/HA scenario. Here’s a pic of the crew that shared in this technology killing session