RSSArcane weekend technical link
Just in case you were looking for something technical to gnaw on this weekend, Henry Newman at EnterpriseStorageForum.com has posted an argument as to why RAID's Days May Be Numbered. It is a bit over my head but if I am reading it right I think he's gone down a rabbit hole relating drive density and hard error rates, a combination which he calculates will begin to cause increasing data loss during array rebuilds. As more data is stored more and more densely, this will become an important factor in enterprise storage system rebuilds.
I find myself agreeing with other comments on the article, though, that if you are using RAID as your ultimate backup and retention mechanism, you are already in trouble; it's a mechanism for enhancing performance and reliability, not for ensuring the indefinite retention of data… although for some applications, it may be quite adequate for that purpose as well. Newman only looks at parity based RAID schemes 3, 5, and 6, and doesn't consider other permutations which avoid the problems he raises. Moreover, the rebuild rate he suggests will result in problems is far in excess of that actually required in most production environments I have seen, and this only serves as a reminder that the Inexpensive part of the name implies a certain failure rate anyway… the real issue is not the ultimate reliability of the system, but the cost/benefit for the reliability level that you can achieve.
Anyway, those of you who are slightly more geeky than myself might wander over and wade into the arguments yourselves. I'm going to breakfast.
