RAID Risks. Actuary for data?

Good read on the numbers behind your storage risk planning.

Personally, when the data at stake is your clients and by inference your company’s reputation and business longevity, nothing short of a 1 in your RAID setup will do. So RAID1 or RAID10. Currently I split data for PostgreSQL on to RAID10 volumes for speed and redundancy with small drives. Main data drives are 300GB 15K SCSI’s at U320 that start in RAID1, then move to RAID10.

I am also a big fan of SCSI bus redundancy in DAS setups. The same as multipath in SAN or iSCSI lingua. Also if you split the mirrors of your RAID1 or RAID10 arrays over the channels, apart from extra redundancy, you also get extra speed.

My only regret is that the Dell PERC4/DC and PERC4e/DC don’t support RAID 51. I have one DAS box that has a RAID 10 array over split channels with 12 drives. 6 mirrored sets stripped across each other.

If I go to RAID50 with one RAID5 per channel, the capacity increases, but the loss of a channel here would be catastrophic.

If I go for two RAID5 arrays, then the loss of a channel is not so bad. However the extra speed is lost and the freedom to address one large block of storage is removed. I would have to manage two data stores. And given some of the maths in the above link, I am not going to do a 12 drive RAID5, again, I must switch off dual channel to reduce the catastrophic failure a failed channel would introduce. So I guess I am stuck with RAID10…..until I replace this DAS with a SAN.

You can’t have it all. Price, speed, capacity, redundancy…..well you can, with RAID10 – but you pay through the nose for it due to the 50% inefficiency of data storage. And unlike RAID5, the efficiency does not improve as you scale. It stays a constant 1:1.