Note there are differences between chips - for example AMD Opteron's all have a memory controller on-board, whilst most Intel processors make do with a single front-side bus. That said I believe the Xeon 55xx series processors are more like Opterons - you may wish to do some reading.

As said though, a single quad-core device, especially if it's Xeon 55xx series, should be more than sufficent. Disk is always the point of contention IMO - better to get the capacity needed with more smaller disks, to get maximum concurrency. To echo the above, battery-backed write cache on your array controller is critical.

Please award points to any useful answer.

Hi Albert

The battery on the raid controller isn't related to other onboard batteries like CMOS - and it's not so much that the BBWC improves the performance, rather that the lack of it destroys performance!

ESX must guarantee to the running VMs that disk writes are committed when that VM thinks they are, so ESX performs no write caching itself. Furthermore all array controllers with onboard cache will run in write-through mode without a battery - this is the issue. For each write command issued by a guest, the array controller must read the stripe, update the data and the parity, and write those back. So each underlying array stripe would have many individual updates applied, destroying write performance - typically we see 3 to 5 MB/s in this mode.

Add the battery, the array controller will flip to write-back mode, since in the event of power interuption, outstanding writes can be committed when power is restored (immediately after the array controller POST). Now ESX passes the writes as before, but they just sit in the controllers cache. Essentially the array controller then only physically commits the writes once it has a complete stripe dirty, or some other time period elapses, so the stripe is hit once in a block write scenario and with no read penalty at all. Then we see write speeds depending on the controller and disks - a simple 3 disk SATA RAID-5 might do 80MB/s.

HTH

Please award points to any useful answer.

Many thanks for the explanation J1mbo.

Cheers,

Kind Regards,
AWT

Make sure the BIOS in the machine supports hardware virtualization if you want to support 64Bit OS. The 800 series processors were towards the beginning of hardware virtualization and depending on the series, may or may not support it. You may be as well off with 3.5 if the hardware doesn't support 64bit VMs.

I would do multiple machines and spread the load.

ESXi has a single LUN capacity limit of just under 2TB so you will need to create multiple arrays or virtual disks depending on your RAID controller.

ESXi does not have support for UPS so you will need to find some way to do an orderly shutdown. Some UPS software can communicate via the network and hybernate. There is also a script that can help shutdown the VM's.

I would do as much testing as you can before even considering going live. You want to understand as much as you can especially with multiple offices.