> > Actually you don't *have* to run fsck at all.  The only thing fsck
> > does is recover lost space.  You can nice fsck down or not run it at
> > all.
> 
> Unless it detects an unexpected error.  I encountered one just a few
> days ago -- duplicate block allocations...  (I'll be posting on that to
> the Xen list.)  On-board disk write caches can easily produce that kind
> of thing if there's a power hit.

By "On-board disk write caches" I assume you mean write caches on the
disk drive.  Those need to be set to write-through mode rather than
write-back mode.

Personally I'd go with nicing fsck down rather than not running it at all.

> (UPS?  Sure; I have one for each of
> my machines.  The machines also have no way of knowing when the UPS
> battery is dying, since last time I checked our USB stack was
> sufficiently incompatible with Linux's that the usual daemons don't
> work that way.)

A UPS doesn't help if someone bumps the power switch.  I've seen that
happen.  Or if a sysadmin makes a mistake and throws a switch.  I've
seen that happen as well.  Or if the UPS just plain fails.  Yep, seen
that one also.

A UPS monitor daemon shouldn't be too hard to write.  Yet another SoC
idea?