[ Date Index ][
Thread Index ]
[ <= Previous by date / thread ] [ Next by date / thread => ]
-----BEGIN PGP SIGNED MESSAGE----- Hash: SHA1 Steve Marvell wrote: > On Sat, Jun 21, 2003 at 02:41:48PM +0100, Robin Cornelius wrote: > > >>PS i wonder what the actual probability is for a double disk disaster? I >>suspose you would have to factor disk life, mean time between failures and >>the actual normal distribution of failure for a given disk which i suspect >>the manufactures would not freely give away. > > > You can work it out with the mean and standard deviation between > failures. The manufacturer normally only quotes the mean, but since > this is time series stuff, it might be a Poisson distribution, which > has sd and mean the same IIRC. Having spent too long doing statistics in the real world, I wouldn't trust a standard deviation as far as I could throw one (how far this would be is left as an exercise to the reader). As HP's guide puts it "the failure rate is assumed to remain constant when MTBF is used to predict performance", so the mathematicially talented can probably calculate a standard deviation but it won't help them much. Thus given a MTBF of 500,000 hours for a disk unit, and an average repair time of 25 hours, given one disk failure, the probability of the second disk failing in that 25 hour period is about 1 in 20,000. This ignores the real world models I referred to earlier, which say irritatings things, like disks are more likely to break when you are replacing the one next to it. I seem to remember a key goal when I was teaching Siemen's hardware engineers Unix system admin skills, was to make sure they could correctly identify from software which disk in a (software) mirrored pair had failed, as hot swapping the wrong disk was a well known cause of unplanned down time! Relying on local system admin skills or hardware failure indicators wasn't 100%. -----BEGIN PGP SIGNATURE----- Comment: Using GnuPG with Mozilla - http://enigmail.mozdev.org iD8DBQE+9sLoGFXfHI9FVgYRAs4cAJ9IrEcVKhYKsmQx6Yx2CNkT1wuaLwCgmwyb CIAjx6VilFYNxD5l9reXerE= =tmJj -----END PGP SIGNATURE----- -- The Mailing List for the Devon & Cornwall LUG Mail majordomo@xxxxxxxxxxxx with "unsubscribe list" in the message body to unsubscribe.