Over on the Citrix XenDesktop 5 forums Citrix says that the VM differencing disks that are used for each VM are NOT aligned and thus cause I/O penalties due to the added physical read/writes. Citrix said this is due to a hypervisor issue, not the way they create the differencing disks.
My understanding is that ESX always creates aligned VMDKs in relation to the physical disk sectors. The only alignment issue one might have comes from the guest OS and making sure within the VMDK it's aligned. This is automatic with server 2008 and later, and requires manual intervention on server 2003.
But that's unrelated to differencing disks, as the guest has no control over them since it's a hypervisor process done in the background. So Citrix says for ESX that you should only use NFS for differencing disks because of the alignment problem. Block storage is discouraged.
ESX/ESXi create aligned VMFS datstore.
But then also the data in the vmdk must be aligned, and this depend by the guest OS.
I realize the guest must be aligned. But I'm talking about the differencing disks. Guests are totally unaware of their existence. Citrix is claiming differencing disks on VMware are not aligned so you should use NFS to reduce IOPS. From what I understand of VMware, I don't think that's the case but I'm looking for a more authortative answer.
Is the differencing disk the same as VMware View's linked clones? I remember hearing from somewhere that the replica disk were not aligned. I can't remember where it was from or haven't found anything to validate it.