Re: Soft Updates use

[Chuck Yerkes <chuck+obsd_(_at_)_2003_(_dot_)_snew_(_dot_)_com>]

Quoting Generic Player (suck_(_at_)_my-balls_(_dot_)_com):
> On Sunday 22 December 2002 09:32 pm, anonymous wrote:
> > Okay, thanks for the link to McKusick's papers on the subject.  I
> > tried to read them but I'm not sure I understood most of the
> > technical explanations. Although they talked much about speed
> > comparisons between journaling and soft updates, I'm still not sure
> > if either methodology is "safer" (or at least less unsafe) than the
> > other in terms of resistance to data loss from, say, power outages or
> > non-mechanical system crashes.  Oh, well . . .
> 
> They are arguably equally safe.  The advantage journalling gives is that 
> you get your machine back up faster, and with the time required to fsck 
> big raid partitions, that can be a big deal.  Theoretically, 
> softupdates will give the same advantage with background fsck, but 
> that's only in freebsd for now.

Except with FFS2, which handles PicoBytes (folks are using
multi-terrabyte files systems and hitting limits).

The leading edge stuff has snapshots.  You crash, it reboots, you
need to fsck, so it gets mounted and works from a snapshot while
the data is fsck'd. Changes are committed later.

Journalling apparently has some advantages (per Seltzer in large
NNTP servers), some journaling file systems let me put the journal
on another file system (including, perhaps, a NVRAM disk) for better
speed.

FFS2 and excessively large file systems were critical for FreeBSD
which is often found in very large web/file server situations.
Not sure about the impetus for the other BSDs - perhaps best to let
the FreeBSD guys burn their fingers and work out the glitches.