Radically different Read latency on same host and lun?

[BradI100]

I have been running a lot of IOmeter tests on my shared storage connected via iSCSi.  I have two different guests 1 is a 2003 server the other is a plain XP Pro workstation.

On the server I get approximately 58 MBps read and write however, on the XP workstation I get 59MBps write but only 4.5MBps read.

They are on the same host.  They are on the same lun.  During testing I shut down one guest before testing the other and vice versa.

The read latency on the XP Pro vm is 10x than the write latency on the XP vm.  I am baffled any hints on where to look for this bottleneck?

Thanks

[mcowger]

They must have different settings in the VM on IOmeter or the OS.  block size especially can do this.

[BradI100]

I ran the same IOmeter tests on both.  But how can I check the block size on my OSs?

Thanks

[thecakeisalie]

If you didn't specify a non-default size when you formatted them, this KB should explain: http://support.microsoft.com/kb/140365

[BradI100]

Thanks.  These are all NTFS formatted drives.  So I guess 4 KB.

[rickardnobel]

Do you test against the system disk or another "clean" vmdk? What kind of performance do you get if just copying some files inside the XP (doing reads and writes)?

[BenConrad]

Are you running Buslogic on the XP box?

http://kb.vmware.com/selfservice/microsites/search.do?language=en_US&cmd=displayKC&externalId=1890

Ben

[BradI100]

Thanks for the replies.  It looks like it was an unrelated firmware error.  I contacted the company (Drobo) and they sent a pre-release firmware that has improved performance significantly.