Joe Chang (jchang6@Xyahoo.com) on 9/8/08 wrote:
---------------------------
>I would be interested in hard analysis on whether an enterprise SSD need be in a
>RAID. The general idea is that the controller on the SSD should be sufficiently
>reliable, only software (or possibly hard) errors on individual memory cells (or
>cell blocks), in which case, this should be handled on the SSD itself. If the SSD
>controller itself were to fail, then it would nice if recovery was available for the good NVRAM devices.


I've been a proponent for flash to replace hard drives for a couple years on RWT. But despite my belief in them, if I switched to an SSD for data I cared about I'd still RAID them. They are still a new technology, and everyone's product is using version 1.0 software and HARDWARE. Human error being what it is, it makes sense being safe.

While traditionally when you mirror drives you use the same model, I think it may make some sense to use different models AND vendors just due to that factor. Well, with Intel's SSDs so much better than the competition, mirroring with one from a different vendor makes it a waste of money from a performance standpoint, but if the speeds of the other brands are 'good enough' this is probably not a bad idea for these early models. I'd hate to have two drives fail at the same time due to some weird corner case for data corruption or failure getting triggered on both at the same time!

I recently bought two new 640GB WD drives to upgrade the storage in my PC, and I was a bit perturbed when they were not identical (one made in Malaysia, the other in Thailand) since they will be mirrored. But as I thought about it I figured that this is actually better, because if there's a problem in one lot or production run of drives, I could be in trouble if both of mine were in that lot. Being made in different places, I'm probably a little bit safer.

I recall at a place I worked a contract 10 years ago their brand EMC Symmetrix 3930 had several drive failures a week for a few months since there was apparently a bad lot of Seagate drives and that Symmetrix had a couple hundred drives from that lot! We didn't lose any data or incur any downtime, of course, but it was still disconcerting seeing the EMC tech there all the time...