Large VMDK and VMFS block sizes or RDM?

[smkelly]

We're starting to see a lot of demand for filesystems that are 250-500GB for servers that contain multimedia files or other forms of storage. One of the things we're working on is moving to a virtualize-first methodology here, so we're trying to accommodate these in VMware before we push them off to physical servers or blades.

My question is whether it is best practice to use larger block sizes that allow for larger VMDKs or to just go to RDMs for these larger VMDKs. I know there are many variables to consider and some of it comes down to preference, but I guess I'm curious what other people are doing out there with these larger filesystems.

For reference, we have 4 ESX 3.5 hosts connected to a Compellent Storage Center over 4Gb/s fiber through Brocade 4900s. We usually use 500GB LUNs with the default block size and put about 16 VMs on each LUN. We have no performance issues to speak of.

Thanks!

[lamw]

We had the same thought too, we had some VMs that had large amount of files and recently we switched raid-controllers and we did not realize that it could not read the headers on the drives and it actually overwritten the old headers. Our data was stored on the large .vmdk and this was a nightmare to get the data back, I don't think we've fully recovered all the data especially not having all the stripe size/header information on the disks, that would of helped more to re-create the header table. I would suggest using an RDM or offload the data to say NFS share or something, with this approach, least the volume can still be mounted up without ESX and it's not stuck on the vmfs volumes. Depending on the mode (virtual or physical) you'll still get some of the capabilities like snapshoting, so it is a personal perference, but when you have user data, it should be stored on external storage like a SAN or NAS device.