On Sat, Jun 26, 1999 at 01:59:19PM -0400, Gregory Maxwell wrote:
> 1 ~30 minute reboot a week is 99.9% uptime, which is what NT hardware
> vendors promise w/ >100k/yr support contracts.. So aparently 33 minutes
> should be perfectly acceptiable to enterprises.
0.998 availability is what ftp.funet.fi has achieved
during last year. It had downtime in 6 occasions,
total accumulated time was around 20 hours. Most of
it was due to equipment move, during which an external
RAID5 controller memory battery fault forced it to
recheck the entire array parities, and that took quite
some time...
(Router problems causing connectivity losses occurred
more frequently and for longer total duration, but
they weren't system problems at the machine.)
For a system like that there is *never* time when
downtime is acceptable. People fetching data are
about at all times; from different parts of the world
at different times, of course. So the only option is
to recover as fast as possible.
That system reboots/remounts its disks at about rate of
about 1 GB / sec. It has 230 GB. Aside of occasional
hardware problems, system boots to full service in about
15 minutes.
Had there been only 7 hours downtime from all those
reboots and moving, the availability ratio would have
been 0.999.
We have considered (at times) using e.g. Linux for the
next generation server, but it won't be happening before
there are truly enterprise class filesystems available
for Linux (or for any other would-be candidate either).
Any non-journaled filesystem is out of the question with
these data volumes. (And still the best FS I have worked
with *anywhere* is DEC UNIX AdvFS. Perhaps the IBM JFS
is second best. Volume management with SGI XFS, and
Solaris VxFS are yet missing experiences.)
/Matti Aarnio <mea@ftp.funet.fi> "play"
<matti.aarnio@sonera.fi> "work"
-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@vger.rutgers.edu
Please read the FAQ at http://www.tux.org/lkml/