[geeks] Fwd: [IP] Interesting speculation on the tech behind gmail
Sandwich Maker
adh at an.bradford.ma.us
Wed Apr 7 15:54:03 CDT 2004
"From: Brian Dunbar <brian.dunbar at plexus.com>
"
"On Apr 7, 2004, at 12:22 PM, Sandwich Maker wrote:
"
"> "From: Mike Meredith <mike at blackhairy.demon.co.uk>
"> "
"> "On Wed, 7 Apr 2004 08:22:46 -0400 (EDT), Sandwich Maker wrote:
"> "> "> The most obvious challenge is the storage. You can't lose
"> people's
"> "> "> email, and you don't want to ever be down, so data has to be
"> "> "> replicated. RAID is no good; when a disk fails, a human needs to
"> "> "> replace the bad disk, or there is risk of data loss if more disks
"> "> "> fail. One imagines the old ENIAC technician running up and down
"> the
"> ">
"> "> this is so glibly wrong.
"> "
"> "Well except in one bit accidentally ... with a large enough number of
"> "disks you will need someone running around replacing the failed ones
"> on
"> "a full-time basis.
">
"> yeah, but -- with 100k cpu boards, they must have someone running
"> around replacing failures already.
"
"CPU board failures happen less often than hard drive failures .. no?
"One imagines there would be less running around.
aren't disk mtbf into the 100000h range now? and - assuming 73G
drives - a 1 petabyte fs would require only ~14000 drives. that works
out to one disk failure every ~7h. with raid5 and hot spares, hardly
a crisis.
mtbf of 100k boards would have to be in the 700000h [the better part
of a century] to be equal. are they really that good?
"From: Lionel Peterson <lionel4287 at verizon.net>
"
"> From: adh at an.bradford.ma.us (Sandwich Maker)
"
"Why replace them as the fail? Wait until you get enough failures in one area, then clear out the broken machines to install known good ones...
that applies to triple-site-redundant storage too.
"At an estimated $300-500/system, I think their idea is that they have all their cold spares "on line" - before a critical number of them "die" they will probably want to replace them with faster, denser, more efficient systems.
that could apply to hot-spared raid arrays too.
________________________________________________________________________
Andrew Hay the genius nature
internet rambler is to see what all have seen
adh at an.bradford.ma.us and think what none thought
More information about the geeks
mailing list