So ESX 3i works on a DELL Poweredge 2650. Period.

At my full time job or as my wife would say my real job, we just retired a couple of older IBM 445 8-way boxes, and replaced them with the X3650 dual quad core with 48 GB of RAM. 4 of them are attached to a DS4400 SAN and our whole production environment is virtualized. Over 100 VMs on those 4 boxes. From our exchange servers to our Oracle 10g databes.

My wife still thinks my side business is just a hobby. I might pick up a Dell 2850 off of EBay for the 64bit capability. I just got my MCITP Enterprise cert and the class got a copy of Server 2008 with 5 cals. So I am itching to get everything moved over.

John

Could that be the speed of your HDD and VM memory over allocated?

Ping Wu

1) No, it was 4 processors on the old machine, 4 on the new VM after the move.

2) In the performance monitor the trend is very close, for guest OS the spikes follow the host OS CPU usage spikes tightly, on all virtual cores.

3) Again, on the graph, the guest and host both show the spikes. Chicken or Egg? It's a sql server that's also accepting a lot of write and read filesystem requests. No other VM's on the machine, this is the only one, no other processes involved. also, I note that when I ping the host adapter, I get a steady and quick reply to each ping, but when I ping the guest IP, the reply times are erratic, and follow the pausing of the mouse and keyboard input, so I suspect it's solely the guest that has the issue.

Thanks for the suggestions. I worked with VMware Tech support on this, and we made some memory resource changes but that didn't solve the issue. What finally corrected it was changing the guest OS's CPU from a Multiprocessor HAL to a Uniprocessor ACPI HAL. Originally, the intent was to "toggle" the HAL from Multi to single, and back to Multi. He had seen this correct similar issues in the past. As soon as we made this change to single core and rebooted, CPU resource consumption dropped from the previous average 75% spikey performance down to 8 - 10%, and performance was back to normal. Since this corrected the issue immediately I didn't bother returning it to multicore.

His theory on this was that while the guest OS was originally using 4 CPU's, the ESX hypervisor and guest were wasting a lot of resources trying to load balance processes that didn't need to be load balanced. By switching to uniproc, you eliminate all of that overhead. By the way, this was a SQL server used for capturing data from a foreign system's "syslogs" so-to-speak, so there are about 15 interface processes capturing data and SQL itself importing it all. Now, the system hums along with resources to spare.