Log in

View Full Version : RAID 5 and RAID 50 reliability


TBone
December 21st, 2004, 18:15
I'm setting up a big RAID array as part of a MSCS cluster for the company I work for (a small engineering company). We're using two Dell powervaults - a 660F chained to a 224F. Each has 14 36GB 10k fibre channel drives.

In the first enclosure:
2 drives in RAID 1 mode will be the cluster quorum (which is overkill, but it requires it's own LUN)
2 drives in RAID 1 mode will be for Exchange (which also requires it's own LUN)

This leaves 24 drives. At least one of these should be set aside as a hot spare for automatic drive rebuilding if one of the active drives fails. So that leaves no more than 23 drives to work with. I want the basically all of the remaining drives to go towards their main storage shares. Read performance is more important that write performance, but neither is critical. The main factor is maximizing storage efficiency while still providing good fault tolerance, and reasonable performance for general purpose storage needs.

RAID 50 sounds like the best candidate for the remaining drives. To me, it seems most logical to make 3 "sub-arrays" of 7 disks in RAID 5, and then raid those three sub-arrays together using RAID 0 into a "super array" of 21 disks. That would leave two more drives unused, but it might not be such a bad idea to have additional spares anyway. If at later date we added another 224Fs with 14 drives, they would be evenly divisble by 7. So this would also work out well for expanding the array at a future date (supposedly Dell's software can do this with the storage on-line if you have the right quantity of disks).

What do you think? There seem to be a lot of different recommendations for how many drives are "safe" to put together in a RAID 5 array. Most people seem to prefer 5-6, and I've heard several recommend no more than 8. Do you think 7 is a good compromise, or is it too big? Also, should I stagger the locations of the drives in each RAID 5 sub-array so that the drives in each sub-array are never adjacent to one another? I'm trying to think of all contingencies, including thermal failure which might effect drives that are physically close to one another. But maybe that's just paranoid

Woodmann
December 21st, 2004, 19:15
Howdy,

RAID 50.
What are you using for controllers ?


Hard drive speed as well as traffic will determine what is best.
You could try up to 14, maybe more. Each array should have its own spare disk.
(the spare disk should equal the smallest "active" disk)

The only way to find out is to try it. I would stagger them.

Then again, you could just ask Dell

Woodmann

TBone
December 23rd, 2004, 00:47
Well, screw it. It turns out that the controllers on the 660F Powervaults don't support RAID 50. In fact, the only concatenated RAID level they support is RAID 1+0. Which is really fault-tolerant, but wastes a lot of space. Now, one might think you could just make a bunch of RAID-5 arrays presented as virtual disks to Windows, and then convert them to dynamic disks and do a striped NTFS volume across the arrays. But no, MSCS doesn't support dynamic disks for reasons that can only be described as "pointless" or "sadistic", I'm not sure which.

So, I'm stuck with doing single-level RAID. And to make matters even more fun, you can only have up to 16 physical disks per disk group, and virtual disks (LUNs) can't span multiple disk groups. So in a nut shell, the largest contiguous drive space I can give them is a 16-disk array of 36 GB drives.

On the plus side, it turns out that it will let you create multiple virtual disks using the same set of disks, even using completely different RAID modes. So, I can take two drives and put both the cluster quorum and the exchange database and logs on the same physical disks without the OS wising up to it. So I only need two disks for that instead of four.

Ultimately what I settled for was this:

The first group has two disks. The first 500 MB of each drive uses RAID-1 to make a virtual disk for the quorum. The remainder of both drives makes another ~35GB RAID-1 virtual drive for our Exchange server's database and logs. The current database is 4GB, but we haven't done any archiving at all. I don't forsee it growing to anywhere near 35 GB until well after we've purchased more storage.

The second group has 16 drives in RAID-5. That's our main storage for active projects, and other general use. It works out to ~500 GB of usable storage.

The second group has 8 drives in RAID-5 (~250 GB). We'll use this one for archiving inactive projects, and I'll probably also put our roaming profiles on it. We might throw other things on it too, but that's for someone else to decide.

There are two spare drives in the system for automatic rebuilds. The disks for all volumes are staggered as well as possible, and they each have an equal number of drives on each channel for better throughput.

As for asking Dell, we didn't buy it from them so they aren't exactly forthcoming. They've retired the powervault line recently and sold out to SMC, which makes the Clariion line now. This is an "on the cheap" setup we got from Ebay. About $3k, including two fibre channel host bus adapters, which run about $250 each, for a refurb. $400+ new. So, not a bad deal all things considered. But the warranty has definitely expired, so I'm on my own for this one.