We're attempting to do Citrix PS 4.5 on Windows 2003 sp2 on ESX 3.01. Currently we run into extremely high context switches, processor queue length and periodically processor utilization %. These Windows metrics/counters usually go take a marked increase at ~6-7 users and user performance dips/become sluggish.
To get a disk i/o account of our current VM performance, I ran iometer against our base Windows 2003 sp2 VM generated from template. I run a 50/50 read-write and 50/50 seq/random test on the VM with defaults in the test beyond that (2 KB transfer, 100% access spec, burstiness, align I/os). I am typically getting back about 300 ios/sec, 0.6 MB/sec, 3.0 io/sec response time. From what i've read, these numbers are slow/low. The iometer numbers are very similar when VMDK files are moved to local SAS drives.
The ESX host is a DL385 G1 with 4 NICs (2 for svc console/VM network, 1 for iSCSI/vmotion, 1 for demo VM network), 2 dual core CPU, 2 internal SAS drives (for ESX), using SW iscsi initiator via Cisco switch to back end NetApp 3020 VMFS volume (500 GB).
2 questions: anything I'm missing iSCSI/infrastructure setup-wise that would help my numbers? are my numbers low/slow?
Do you have any A/V running? Any real-time scanning would skew IOmeter results by huge amounts (i know, I accidently did it once!).
Dave
Very interested in how your implementaton has progresed. We are running a near identical config and are experiencing huge performance issues with Citrix. We are having to add additional Citrix VM's, now about 8, to handle a farm of 250 users. This is not what I expected. We have reduced the processors for each VM to 1 as reccommended but performance is deplorable. Any specific settings for the VM or host that would make a difference?