Jeffrey Hutzelman <[EMAIL PROTECTED]> wrote:
}Well, I can tell you how we're set up...
Which I will do as well.
}We currently have on the order of 40 fileservers, plus 3 dedicated dbservers.
}The database servers and most of the fileservers are all sparc-1 class
}machines, though we've been slowly upgrading the fileservers (we'd go faster,
}but there are software considerations involved as well). When we're done,
}we'll have some smaller number of Sparc 4/5 class fileservers, and the
}same 3 Sparc 1+'s as dbservers; we've never seen a preformance problem
}because of this.
+ We have 3 database servers (bottom-end Alphastations)
Once you get to any significant size, separating these
is a big win, and you don't need a killer box for them,
just scrounge up what you can. Ethernet-switch attached.
+ We have 5 ReadOnly fileservers (DECstation 5000/240s)
which serve /usr and are not backed up (these date
from the days when a disk big enough to hold /usr
cost more than the rest of the machine). 2 are
ethernet-switch attached, 3 are FDDI attached.
+ We have 11 ReadWrite fileservers (10 DECstation 5000/240s,
1 Alpha 3000/600) serving /home which are backed up weekly
(full backup) with an attached 4mm DAT drive. Approx 250GB.
FDDI attached.
+ The 10 ReadWrite DECstations above are being replaced by
3 AlphaServer 1000a's with RAID. These will be backed up
weekly (full) and nightly (incr) with attached 40GB/tape
DLT drives. The Alpha 3000/600 above will also follow
this backup regime. FDDI attached.
}We have about 500GB of data online; most but not all of that gets backed
}up; unfortunately, I don't have exact numbers at the moment. Backups
}are handled by a DECstation 5000 (that's a MIPS-based box, not an alpha),
}with a single 8mm drive for backups and a separate drive for restores
}and as a spare. We do fairly aggressive backups; each volume gets a
}full dump every 6 months, and incrementals every month, week, and day.
}We're just now approaching the limit of what that single system can
}do in a day; with any luck, the next batch of improvements to our backup
}system will both improve the performance of the existing system, and make
}it possible to add additional drives and/or machines. Right now, the
}performance bottleneck is not the machine; it's the tape drive and, in
}some cases, the design of the backup system we use.
It sounds like you are doing backups over the net --
this is something we've never been able to get to perform
adequately... BTW, the DLT drives we have are lots faster
than the 4mm we've been using -- we can backup 75GB in
an 8hr shift through AFS, about 10x what we're getting
with the 4mm (obviously a little more CPU power isn't
hurting this either, but...).
}So, I'd reccommend improving your fileservers first. Of course, in your
}setup (as in many others), the dbservers are also fileservers, so that
}means they'll be upgraded too. With a small cell, that's probably OK;
}in a larger cell, dedicated database servers can both improve performance
}and make things easier to manage (e.g. you're never taking down a DBserver
}to add or replace a disk). Of course, YMMV.
The major cause of downtime for us is also adding/removing
disks -- it would be a win if the fileservers could notice
(or be told to nocice) a newly arrived disk... (*dreaming*)
John
--
John Hascall, Software Engr. Shut up, be happy. The conveniences you
ISU Computation Center demanded are now mandatory. -Jello Biafra
[EMAIL PROTECTED]
http://www.cc.iastate.edu/staff/systems/john/welcome.html <-- the usual crud
