VSA HDD planning

[oikjn]

Has anybody figured out what the ideal setup should be for the software VSA hdd setups?  I know that the VSA software does raid 0 across any disks presented...  if I have a set of disks on a raid controller presented to the host as a single drive, would it be best to create one large drive to present the VSA or a couple smaller drives to present the VSA.  I'm thinking one large might be better because it will cause less random IO on the actual raid group, but I wanted to see if anybody has played around with this in practice. 

I would rather use a few small disks because then it gives me the flexability of more easily moving around my VSAs if needed, but as I think about this issue more, I'm thinking I might have to give up on that idea and either go with one large disk, or maybe even directly present the raid disk to the VSA.

[Paul Hutchings]

I'll be very interested to know the answer to this one, if indeed there is a "right" answer rather than opinions.

I'd taken the view that since the P4000 virtualises storage into a single "pool", it makes sense for the underlying disk to be a single pool too, so basically it should come down to picking a RAID type and going with a pool big enough for a particular VSA to reside on it.

Of course I'm making a few assumptions like no 15+1 RAID5's :)

[5y53ng]

I always carved my physical disks into multiple virtual disks on RAID 5 given the 2TB datastore limit in older versions of VMware. Since the VSA has a 10TB storage limit and a maximum of 5 virtual disks (if I recall) I ended up with 5 virtual disks of apprximately 1.9TB attached to each VSA.

I can't say for sure this is the best approach, but given the constraints we were working with in previous versions of VMware/SANiQ this is what I ended up with.

[oikjn]

yes we are limited to 5 disks of 2TB each per VSA.  ESX might have had that limit, but I think its also a limit for HP since I believe Hyper-v doesn't have the same size limit, but the VSA still holds the 2TB limit.

Where I was torn was if it was better to have 4 500GB disks or one 2TB disk if you only want a total of 2TB.  I WAS thinking that the 500GB was the way to go even if all four disks were on the same physical raid group, but I'm really second guessing myself on that now.

If I go with one large disk to the VM the question becomes, do I use VHDs or do I present the disk directly?  I"m leaning towards direct disks presented to the VSA... not so much becase of the VERY SLIGHT performance benefit, but because if I use a .vhd I need to make sure the drive containing the .vhd has some free space where if I just present the drive I don't have to worry about that...  I also don't have to worry about messing with cluster sizes of ntfs that best works for the VSA when I don't know for sure what that cluster size is. 

I have a dozen 750gb drives that I'm going to plit into two six group raid10s which should come just about to 2TB each.  I thought about doing one large group of disks and then using the raid controller to present two 2TB disks instead of one 4TB disk, but the more I think about it, if the VSA is going to span the two drives I shouldn't make a span, split it to have it spanned again (confusing!) and I should just keep the two disk groups seporate so that each is only dealing with one stream of random IO instead of always having to deal with two streams of random IO but with a pool that is twice as large.

I guess I'll find out if this is better since I"m going to add two VSAs using this new method and adding them to a cluster using a bunch of small .vhds.  Whichever becomes the bottleneck will determine which is the best way to go. 

I'll report back my results here, but given the time it take for restripes of TBs, this will be a while... that and I don't have the physical new computers hosting the new VSAs yet.