virtual disk perf question - IDE vs LSI vs Paravirt

[TedH256]

Just a poll of opinions:

A customer has several distributed stand-alone ESXi 4.1/ 5 hosts with 8+1 raid 5 local disk arrays, running virtual machines that were V2V’d from Hyper-V hosts. These are primarily W2k3 servers running an Progress DB-based application. The virtual disks for these VMs are IDE (I’m not sure if they were IDE disks on Hyper-V or if the VMware converter changed them to IDE).

Do you believe that a large performance gain could be expected if the data drives on these machines were converted to either LSI or the paravirtualized SCSI controller? This hosts CPU and Memory resources are not fully utilized and do not appear to be the cause of perceived poor performance.

[harrygunter]

I would change to paravirtual controllers myself, make sure you have the lastest tools installed.

As they are 2003 guests i would also check the disk alignment on the drives, 2003 by default does not align the disks unlike 2008.

Mark

[TedH256]

excellent point re: alignment!

[rickardnobel]

TedH256 wrote:

Do you believe that a large performance gain could be expected if the data drives on these machines were converted to either LSI or the paravirtualized SCSI controller?

You could install a similar Windows 2003 VM with LSI Logic / PVSCSI controllers and run a disk performance test tool on this and on the converted IDE systems and see if there is any gain in this configuration.

[jr195]

According to a VMware storage presentation at the Silicon Valley VMUG a few weeks ago, VMware now fully recommends using the paravirtual drivers. Apparently the older issues with them not being optimized for low i/o VMs or suitable for a boot disk have been fixed.

As another post said, though, the paravirtualized controllers are going to lessen the CPU load and time processing i/o requests, and is not going to speed up your underlying storage