TR Forums

I've got access to Dual Quad Xeon machines running VMWARE ESX. Unlike VMware server which is limited to 2 cores / guest, ESX allows 4 cores / Guest with the ability to set guest affinity within ESX.

I have 2 Ubuntu guests running with the affinity in guest1 set to cores 0-3, and guest2 set to cores 4-7.

ESX is reporting that the 8 CPU cores from the 2 guests are only 70% active, and 30% of the cores are idle.

Unfortunately, adding a third client resulted in the original 2 slowing down, and the net effect was a jump to 80% CPU use. I assume that the 4th guest will utilize 95% just like VMware Server requires 2 guests to max it out.

I hate to have to run 4 2-Core ESX Guests. Anyone have experience running on ESX??

Ok, maybe spell out what you have a little better.

You have two quad Xeon machines. Which Xeons? While I'm guessing some of the newer Core-based Xeons, you could have P4 Xeons or even P3 Xeons for all I know. The logistics of each situation is a little different.

From the work I've done with Hyper-V, ESX, and GSX (before it became Server) and more recently Server 1.0 and 2.0, with primarily Linux VMs, you'll see Linux VMs easily scale from 1 to 2 cores, but beyond a few specialized scenarios, it doesn't scale well to >2 cores. The Windows Server VMs I've worked with scaled even worse, though.


What you really need to ask here is 'WHY'. If you're just looking to see the CPUs pegged out, look to your workloads. Find the bottlenecks *inside* the VMs. On the other hand, if you'd just be happier seeing the CPU usage pegged out, just install F@H and forget it. If you're looking for maximum performance, you're probably as close as you'll get when you have the 2 VMs and 70% utilization.

CPU utilization isn't everything. Look at your ram and disk load levels, too. If they peg out early, no number of additional VMs will give more work.

I suspect that you may have responded without realizing that this was posted in the Folding forum, so my comments were specific to running the SMP Folding client on ESX.

They are dual socket Xeon E5430 2.65Ghz machines, so 2 CPUs x 4 Cores=8 Cores. There's currently no other load on the machines, so I'm running a CPU burn "test" with FAH

Currently the machines have downloaded 2665 WU's and are down from 70% to 55% CPU utilization. Memory and Disk utilization is negligible.

I think you've answered my question with your statement Linux "doesn't scale well >2 cores". So, I'm back to running 4 FAH Clients to max burn the CPU's.

JPinTO wrote:

I think you've answered my question with your statement Linux "doesn't scale well >2 cores". So, I'm back to running 4 FAH Clients to max burn the CPU's.

4 SMP Linux clients? That will be one heck of a folding monster.

[quote="Flying Fox"][quote="JPinTO"]I think you've answered my question with your statement Linux "doesn't scale well >2 cores". So, I'm back to running 4 FAH Clients to max burn the CPU's.[/quote]
4 SMP Linux clients? That will be one heck of a folding monster. [/quote]

It still will be bested in PPD by a Quad Core + 8800GT combo.

JPinTO wrote:

I suspect that you may have responded without realizing that this was posted in the Folding forum, so my comments were specific to running the SMP Folding client on ESX.

Doh. Yeah, I just happened to catch it on the front page in hot topics, assumed it was about maximizing ESX performance, didn't even see which forum it was in.