MickMcLovin
Contributor
Contributor

ESX 4 - Expected Performce...

After attending several Vsphere briefings and reading various articles on benchmark performance increases in Vsphere over ESX 3.5 there seems to be alot of comflicting figures as to how much of an increase we'll see.

In one cluster the hosts have 32Gb RAM and many of the VMs have 12Gb and 4Gb allocated so the hosts only have 2 VMs running. My major concern is at one briefing we were advised that unless running approx 10 VMs on a host we wouldn't see much of a step up in performance??

Just wondering how true or untrue this is??

Tags (2)
0 Kudos
3 Replies
Josh26
Virtuoso
Virtuoso

WIth two VMs under ESXi 3.5 I'd expect to see near physical performance anyway.

I'm sure there are VMWare sales figures out there showing a marked improvement, but realistically I'd struggle to see it being that big an improvement.

0 Kudos
MickMcLovin
Contributor
Contributor

So would you expect better performance from a host with more physical RAM running more VMs or the same host with less RAM running less VMS?? Whether thats 3.5 or 4??

0 Kudos
Rumple
Virtuoso
Virtuoso

There is no magic number, especially when you appear to be running VM's that mimic phyical hardware. Overall the memory addressing, IOPS limits and particular VM limits have been raised, if your current VM's are overprovisioned on resources then you are not going to see any increase in performance.

Most VM's I run are started out with a single CPU and 512MB of RAM and out of 50 odd servers in each cluster i have about 5 that have either additional CPU or additional RAM and nothing is above 2GB and that includes servers like SQL, oracle and BES....

If you are running VM's with 12GB of RAM allocated to them and its using all of it (by checking vCenter not windows task manager) and you are not getting anywhere near the IOPS out of your storage then its capable of providing (based upon the disk and lun design) then you should see an increase in performance.

As for CPU performance, there will be no difference in individual CPU performance as thats really dictated by the hardware. On an oversubscribed server where there is CPU contention at the moment, you may see some performance increases if you are using mostly single CPU servers. If you have 8 cores and 4 - 2 CPU VM's, you are going to be SOL in the performance area... the scheduling required to get 4 vCPU's off pCPU's is still gonig to be a problem...