I'm currently running ESXi 4.1u1 Essentials Plus (no storage vMotion) with two hosts (soon to be three). Storage is a XIO ISE2 array, 8Gb FC. My current datastore is a 2TB LUN formatted VMFS with an 8MB block size. So far with 12 guest OSs we hardly work the storage at all. Our backup is disk to disk (using the ethernet network).
I'm migrating data (network shares, Windows user profiles) from our existing file server (physical W2K8) to a virtual W2K8R2 instance - right now the server is using about 1.2TB of storage on our older Xiotech SAN (broken into 3 LUNs).
The question I have is: am I better off (from an IO, backup, management, etc.) creating a separate datastore for this one VM of 2TB and calling it a day (then creating vDisks within that datastore) or creating an extent and extending the current datastore into it? I want to keep things as simple as possible; however, if extending to another 2TB LUN is going to create contention on the storage or do other strange things, then maybe my best bet is to just create a new datastore specifically for this server?
Finally, the other major consideration with adding an extent (again, I'm on 4.1u1) is whether or not I'm going to have to take down the whole environment to do the extend?
Thank you very much for your time and insight!
Even though an extent can be added on the fly, I would recommend against it. IMO this adds additional complexity when it comes to host version upgrades and there's no real benefit, because the virtual disks cannot become larger than the limit (2TB minus 512 Bytes) anyway. Adding a second LUN/datastore would be my preferred way to handle this. I don't know your storage system, but if you have multiple managing controllers you could benefit from distributing traffic with separate datastores. In addition, with ESXi 5.0 you don't need extents anymore since LUNs/datastores can be up to 64TB.
This is kind of the way I'm leaning as well. I don't see us going to 5 any time soon so we're kind of stuck with the extent method. The storage has two storage controllers that talk to the physical disks. Each controller has 4 ports for a total of 8 FC ports to run traffic through. I have to manually load balance (through zoning on the FC switch) which storage ports the ESX host server uses.
I guess I'm just trying to figure out why I would extend the datastore? Our backup solution works on the OS level (agents) so it doesn't even know about datastores. I'm not sure of the advantages? ESXi 5 would def. be nice!