I've been testing Server 2008 SP2 (not R2) x32 and x64 performance under ESX 4.1 and I have found that the performance of x32 just simply sucks.
I use a simple program that writes times the write of a big file, then times the read, and with the default settings, I can't get x32 to write faster than about 11MB/second even to a ramdisk in the virtual. With X64 on the same host and datastore, I can ten times that write rate to the virtual's C: drive.
Mapped drive performance sucks as well, never topping 11MB/second. X64, on the other hand, gives me rates at about 60MB/second to the same share.
Does anyone have a fix for this?
A little google-fu turns up the following lines to add to the VMX file:
MemTrimRate = "0"
sched.mem.pshare.enable = "FALSE"
MemAllowAutoScaleDown = "FALSE"
mainMem.useNamedFile = "FALSE"
prefvmx.minVmMemPct = "100"
This does seem to at least double the performance of Server 2008 SP2 x32 virtuals, but I'm not sure I like the host performance hit that this would entail. Also, I don't understand what sched.mem.pshare.enable = "FALSE" does on ESX. On hosted VMWare server products, it tells VMWare to write the virtual's memory file on /tmp instead of in the same folder as the virtual's files. I don't think this applies for ESX. As proof, I vmotioned a virtual that had these lines in the VMX file. If the virtual had a file open on the local host's /tmp folder, the vmotion would not have worked.
Ideas?
