Background fsck

Chris Pressey cpressey at catseye.mine.nu
Mon Jan 19 12:44:32 PST 2004


On Mon, 19 Jan 2004 21:14:41 +0100
Rahul Siddharthan <rsidd at xxxxxxxxx> wrote:

> Quoting Chris Pressey <cpressey at xxxxxxxxxxxxxxx>:
> > My own opinion is that the "speed-up-recovery-after-a-crash" thing,
> > while apparently a very sexy problem for filesystem programmers to
> > tackle, is going to become more and more of a non-problem as UPS's
> > become more and more consumer-friendly.
> > 
> > Like, if I can get one of these:
> > 
> >   http://www.outletpc.com/c5401.html
> > 
> > for $20US, and if it gives me enough time to sync everything to disc
> > before the power really goes out, then my HDDs will rarely power
> > down in an inconsistent state anyway, yes?
> 
> Well, power failures are only one possible cause.  System
> crashes/lockups, either because it's an "unstable" development
> platform or because of buggy hardware or device drivers, are (for me
> anyway) a much more common cause of unclean disks.  And on FreeBSD
> 5.x, if you have a linux ext2fs slice mounted (even if it's read-only)
> and you don't explicitly unmount it before shutting down, the buffers
> don't sync.

I grant you that power failures aren't the only cause - but they are the
only cause that no amount of software engineering can prevent.

If the effort that goes into designing and writing new journalled file
systems was instead put into making sure the system itself didn't ever
put the discs in an inconsistent state... well, I wonder, is all.
Suffice it to say that I don't consider background fsck to be much of a
priority.

On the other hand, I'd be happy to see a new FS in DFly, and if
journalling gives the sysop some benefit besides merely faster recovery
times - even if that benefit is "only" that the code is no longer
"guru-dependent" - then I'm all for it.

-Chris





More information about the Kernel mailing list