[3065] in SIPB-AFS-requests
Re: FY98 budgeting and AFS
daemon@ATHENA.MIT.EDU (John Hawkinson)
Wed Jul 15 22:26:07 1998
Date: Wed, 15 Jul 1998 22:25:53 -0400
To: sipb-afsreq@MIT.EDU
In-Reply-To: "[3059] in SIPB-AFS-requests"
From: John Hawkinson <jhawk@MIT.EDU>
[ Dropping sipb-machine-room ]
| I mainly want to talk about AFS right now. ASO has recently qualified
| some 72GB Box Hill RAID units, which cost about $23,000 apiece. I
| think it would make our cell easier to manage if we started to
| transition to these units instead of maintaining a plethora of small,
| differently-sized disks. And, of course, RAID units would make the
| cell a lot more reliable.
I think it's admirable that in the past we've been able to avoid
putting all our eggs in one basket, at least with AFS.
I dislike the concept of a single monolithic disk attached to a single
server. I'd be significantly more comfortable with, say, multiple 20gb
RAIDs.
I don't think that the current level of managability of our cell
has exceeded the bounds of our ability to deal with.
I'd be a little concerned that the RAID may introduce unfortunate software
dependancies ("No, you can't run Solaris 2.7, you have to wait
for the RAID vendor to upgrade their software") that might prove suboptimal.
Despite my arguments in other planes, I think it's important to see how
ops does rolling this out before we move to it. I think we'd like to
consider waiting for AFS to become a bit more mature with regard to huge
disks (fileserver attaching volumes, etc., etc.).
On the other hand, I don't think the issues of failure signaling are
all that relevent. I believe that the box hill, and any other vendors
we would consider, all syslog, or potentially do more interesting things,
but notification definitely happens.
I think that comments about rocket science and statements of "shock,
nay even dismay" (not to be confused with "shock, nay, even dismay")
should probably be kept to a minimum, especially to the extent that they
cast aspersions on members of the Aero/Astro community.
Frankly, neither a menu system nor a perl library feels very comfortable
to me. Sure would scare me.
Restoring a 72GB unit is a Big Production. For the moment ignoring the
widespread lack of backup restoral procedures and knowledge in the
SIPB, restoring a unit of that size is going to take A Lot Longer.
I suppose this is a generic argument in favor of reducing the cell
to a 1.44mb floppy, so shouldn't be leant to much credence.
I concur with Matt that fewer people are likely to be RAID-familiar than
Seagate-disk-familiar. I don't think this is a prohibitive, or even strong
reason, not to consider the RAID.
--jhawk
