In a meeting w/ friends, I was advocating the merits of the home server w/ RAID.
ctlee argued that a RAID 1 (mirrored) set relies on a relatively weak assumption that both drives would not fail at the same time. Whether the assumption is “weak” maybe arguable, but after my home server having survived a HD failure, I am confident that this assumption is good enough. I think ctlee might have confused the probability of any of a collection of parts failing (sum of probability of each part failing, which adds up rather quickly) w/ that of all parts failing at the same time (product of probability of each part failing, which tends to approach zero).
I also suggested that it would be a good idea to use HDs of different makes and/or different manufacturing dates in a RAID set, which ctlee also didn’t quite agree. The logic behind this is based on the assumption that HDs of the same batch (i.e. same manufacturer, very close manufacturing dates) have very similar manufacture characteristics and profiles, so given that they will be in the same RAID set, they will likely fail at about the same time. When you run a RAID, you do not want your HDs to all fail at the same time, and using HDs from different makes should avoid this. A couple of years ago IBM made a big batch of Deathstar Deskstar HDs which were all quite unreliable. A RAID formed by a couple of these Deskstar could be disasterous. Granted Deskstars are cheap IDE drives and seldom used in RAID sets, but you get the idea. This complicates somewhat the procurement process for large enterprises so may not be the norm but to me it makes perfect sense.
1 response so far ↓
1 James // Dec 8, 2003 at 8:19 am
CTLee disagree with everything that everyone else said before he did. If you go back to him another day and say the reverse of what you have said which he had disagreed, he would disagree that as well implying his agreement on what you said first, unless you repeat what you have said first afterwards.
In this case, at least he was being consistant. Had he not insisted on the second part of having the same type, make and batch of HD used in a RAID mirroring, it would have greatly reduse the chance of his first arguement of both HD failing at the same time.
But if you go back to him and say you would not RAID mirror at any cost on another day, he would say you must and must use two different HD – maybe even different size also; not in light of having corrected by your first argument, but just to oppose your new arguement.