[3062] in SIPB-AFS-requests
Re: FY98 budgeting and AFS
daemon@ATHENA.MIT.EDU (Bill Sommerfeld)
Wed Jul 15 18:57:38 1998
To: Greg Hudson <ghudson@MIT.EDU>
Cc: mhpower@MIT.EDU, sipb-machine-room@MIT.EDU, sipb-afsreq@MIT.EDU
In-Reply-To: Message from Greg Hudson <ghudson@MIT.EDU>
of "Wed, 15 Jul 1998 18:15:49 EDT." <199807152215.SAA11877@x15-cruise-basselope.mit.edu>
Date: Wed, 15 Jul 1998 18:56:09 -0400
From: Bill Sommerfeld <sommerfeld@orchard.arlington.ma.us>
I don't understand this argument at all. Why does a Box Hill RAID
unit require "Box Hill RAID familiarity" but a Seagate disk does not
require "Seagate disk familiarity?" I guess you need to know how to
deal with individual disk failures in a different way, but I don't
expect there's any magic there.
Actually, my gut feel is that there's a whole lot more `magic'
involved vs an individual disk..
I don't know anything about the Box Hill unit, but the one high-end
RAID unit i'm familiar with (a Data General RAID unit remarketed by HP
under a name I now forget) had dual-redundant everything, (including
two control processors and two serial console ports), and a user
interface for configuring `slices' of the RAID in various modes. (you
could also configure it over the SCSI bus using SAM from HP-UX).
there was much complexity.. it was a fun toy to play with when we were
setting it up, which is not something I'd say about your joe average
fast-wide SCSI drive..
Now, dual-redundancy is nice, because it means you survive outages,
but if you aren't paying attention to it, Murphy says that one side
will fail, and then the other side will fail a month later before you
notice and correct the first failure. If ASO has infrastructure or
tools which we can share for detecting partial failures and alerting
operators, great, but if the strategy is "patrol machine rooms
sporadically looking for red lights on the disk array", I think we've
got a problem..
- Bill
