Unsolved
This post is more than 5 years old
43 Posts
0
756
Max number of disks in a Virtual Pool
Is there a best practice for the max # of disks in a virtual pool? We have a virtual pool that is spread out over 800 x 450GB disks (RAID5 3+1). Should we be concerned with such a large fault domain?
KeystoneCruiser
22 Posts
0
November 7th, 2011 13:00
We have a very similar situation and I would be very interested in hearing others thoughts on this. Our EMC representatives have said you can grow the pool as large as you want but where do people think it becomes to big?
Quincy561
1.3K Posts
0
November 7th, 2011 14:00
The larger the number of disks in the pool, the greater the chance the pool will experience more than one disk failure in a raid group. The size of the drives is a factor since the larger drives take longer to rebuild. The best practice document talks about this, but does not specify any limit. There are lots of factors to take into account.
RobertDudley
448 Posts
0
November 7th, 2011 14:00
No you shouldnt be concerned about a disk failure. Spares are not permanent spares so on a failure the failing disk is permanently swapped out with an existing spare. You would have to have two disk drives hard fail at the same time that your thin volume spanned to have an issue.
Allen Ward
2.1K Posts
0
November 7th, 2011 19:00
I don't disagree that the risk increases with a greater number of drives involved, but if you are using RAID5(3+1) as an example, the concern isn't based on the number of drives themselves, but the number of RAID5 sets involved. You don't just have to have two drives in the entire pool fail at the same time (with 800 drives in a pool this wouldn't be an unreasonable thing to happen). You would have to have two drives out of a specific four drive combination fail at the same time. It's a bit late for me to be doing the math on this, but considering the low rate of occurrence of double drive faults on an array in the first place the difference in risk goes from infinitesimal (for a very small pool) to almost infinitesimal (on a significantly larger pool) all the way up to downright unlikely (if you made a single pool using the full 2400 drives possible in a maxed out array).
I would be curious to see if anyone has actually done the math on this at some point and come up with some real numbers to put around this risk.