SQL Server Performance Forum – Threads Archive

RAID Level Confusion

I’m reading Kalen Delaney’s Book "Inside SQL Server 2000". She defines RAID 10 as, "two stripes of ten disks each and then mirror one stripe on the other". She defines RAID 0+1 as, "creating a stripe on ten mirrored volumes" She also claims that a RAID 0+1 is much more fault tolerant than a RAID 10 array. (page 135-136 for anyone that has the book) Doesn’t she have these backwards?

From the Microsoft SQL Server 2000 Failover Clustering white paper: "Data drives should use the RAID configuration of striped mirrors for maximum availability. This means that first the drives are each mirrored and then the whole thing is striped. Striped mirrors are sometimes known as RAID 1+0. The next best for availability, which also gives a little better performance than striped mirrors, is mirrored stripes, which is sometimes known as RAID 0+1. Mirrored stripes mean that a group of disks is striped as one set, and then mirrored. It cannot tolerate failures as well as striped mirrors. Note: The terminology for striped mirrors and mirrored stripes (i.e. 0+1, 1+0) may vary from manufacturer to manufacturer." Notice the last comment [<img src=’/community/emoticons/emotion-2.gif’ alt=’

‘ />]. In other words people say either RAID 1+0, RAID 0+1, RAID 10 etc. when they in 99% of the cases mean striped mirrors. Doing the other way around, by mirroring stripes is just stupid in my eyes. If you have 10 disks its much safer to mirror them in pairs and stripe the mirrors, than making two stripes of 5 disks and mirroring them. When you you mirror stripes you can only handle the loss of one disk. /Argyle

Thanks for the clarification. My understanding was the first number was the raid level applied first. Raid 0+1 – Striped and then mirrored.
Raid 1+0 (or RAID 10) – Mirrored and then striped. Most of the documentation I have found indicates that RAID 0+1 is not suitable for Database applications. Don

]]>