In my production datacenter, I am running 3 Dell PE R610 hosts, with 192GB RAM each and six core procs (dual).
Our total VMs across the ESXi cluster is @150 with approximately 70%-ish utilization.
Tim,
In my production datacenter I am running HP DL385 G7 Series Servers with 12 Core Opterons, 192Gb RAM, ESXi 4.1 and average 30-40 VM's with power to spare. It really comes down to density, how may VM's can you get per core and what will it cost. The DL385 is a 2 socket design which can provide you cost savings on licensing over a 4 socket design. The 12 core processors provide enough horsepower for good density numbers. My host CPU averages 8%-11% during off hours and 11%-22% during peak hours. We performed density testing on several models before we settled on our spec. For testing we created a load test template which we used for rapid deployment on each platform using W2003R2 with 2CPU, 2GB RAM and a script which loaded the CPU's and compensated for differences in physical processor speeds. One day I will see how many VM's it takes to get my host CPU utilization up to 80% range!
Mark
~ 100 Server Class (not VDI) VMs per host, wild mix of OSes, a bit of everything
18 hosts in a high density, high performance design
Well tuned SAN / Storage multi-channel configurations
IBM x3850 X2 / 256GB RAM / 6 cores x 4 sockets
~ 70-80% pRAM utilization, slightly less for pCPU
I do not think 8 socket boxes are the best way to go today with increasing core counts. The added complexity of getting there, witht he added latency across those CPU links and possibily halving your server count yields a larger impact on failure.
For scaling up further to 200-300 Vms per host, , I prefer the 4 socket solutions until I see the pCPU run out of steam.
Cheers, G. Mobley
-
vCell03.jpg (109.2 K)
-
vCell03A.jpg (91.4 K)
Actions
More Like This
- Retrieving data ...



