Subject: Re: Crash-resilience of FFS (w/softdeps)
To: B. James Phillippe <bryanxms@ecst.csuchico.edu>
From: Matthias Buelow <mkb@mukappabeta.de>
List: netbsd-users
Date: 11/29/2001 12:23:43
B. James Phillippe writes:

>I'm a NetBSD newbie running NetBSD-1.5.2 on an Alpha with /usr mounted
>using softdeps.  I'm curious to know how resilient the FFS is, particularly
>when using softdeps, on unclean shutdowns.  I come from a heavy Linux (and
>therefore EXT2) background.

Softdeps is principally safer than full async mode at the same
speed but since it is rather buggy on NetBSD it is itself the cause
of about all panics I've had since 1.5.  You'd probably want to go
with the slower performance of the standard settings (async data,
sync metadata) which is only really slower when doing heavy metadata
operations (mostly inode and directory updates) than with the
fragility of the softdeps code in NetBSD (interestingly, softupdates
on FreeBSD doesn't seem to be as instable although it's the same
thing, I hope the remaining bugs will get fixed in NetBSD really
soon.)  For now, if you're interested in stable operations, don't
enable softdeps.  The issue might have been relieved in -current
(I only have softdeps experience with 1.5.x) but OTOH, I haven't
seen any spectacular "massive bugs squished in softdeps in -current"
message on mailinglists either.
FFS in standard settings is rather stable concerning the integrity
of the on-disk filesystem on crashes, although, given enough bad
luck, you can always end up in a situation where the filesystem is
messed up beyond what fsck can repair, also with journaled fs
implementations, or whatever.

--mkb