raw data mappings vs standard VMFS in windows

[ranj]

Hi

I am confused and wanted to get some opinions on this topic.

Currently in the process of migrating an existing physical file server onto a virtual machine in ESX4.0. This servers primary role is to manage home directories for all users and roaming profiles for all users .

I unsure once migrated to a VM which storage topology would give me the best performance VM as a VMFS volume or to use raw data mappings.

Would I be better importing it as a VM and then the VM will be in a VMFS storage type or creating a new VM but creating an extra local disk and using raw data mappings so this disk is accessed directly to the SAN?

I believe the second option would give the best performance but am unsure.

Are other techs in a similar position, can one recommend one over the other? I assume also option 1 would be more easier to manage also.

Any advice would be greatly appreciated.

[CyberNerd]

When I migrated our fileservers, I put the data onto a separate ext3 partition on the SAN that was a raw mapping, but used the virtual o/s as VMFS.

The theory was that it if the user data was on a SAN as a raw mapping I would have better chance in recovering/remapping if something when wrong. If I had better training in ESX then I may have made a different choice, but this currently works for us.

[DMcCoy]

I only use vmfs with vmdk disk images. The overhead is fairly minimal for vmfs and it's useful for snapshots, exports, vcb and svmotion.

There is only one warning with virtual disks, don't revert a snapshot if you added an
*existing* disk image to it after it was taken, the revert deletes the added disk image (oops!, luckily I had a backup from the night before and it was a holiday).

Apart from that I've had no issues.

[Theblacksheep]

vmfs here too. For everything.

RDM is recommended for large DBs... The scale they mean by large DBs and the disk IO vsphere can take will be off normal school charts, so it'll be more depend to your harware (iscsi, fc) and the current setup you are trying to duplicate in the VM world.

For instance, say the file server in question is currently one big server with all files spread across one big space on a 5 disk raid then in the VM it maybe better to split the larger disk into a few smaller ones and place them across several luns.

Convert and test.