Re: Removing softdep

[David Holland <dholland-tech%netbsd.org@localhost>]

On Mon, Jun 09, 2008 at 09:19:19PM +0100, Andrew Doran wrote:
 > > > I also
 > > > know from experience that softdep on NetBSD does not fulfill its data
 > > > integrity promises.
 > > 
 > > Many people have had little trouble with it, though. For me its record
 > > has been noticeably better than ffs *without* softupdates.
 > 
 > I spent a couple of months over the past year doing little else other than
 > trying to break the file system code so that I could fix the bugs exposed.
 > My experience doesn't tally with yours, so we will just have to disagree on
 > this one. Here's hoping you don't run into Murphy any time soon.

Yes, well, normal operation doesn't consist of trying to break the fs
code. All I can tell you is what I've seen, and based on what I've
seen the exposure under normal operation given the workloads I tend to
run is less with softdep.

I particularly dislike silent data corruption.

I would like to see a fs that can be pounded on safely and that
reliably survived power failures while being pounded on. We don't
currently have that. I'm prepared to put in the time once I finish
with certain vfs issues that are basically prerequisites.

There are various ways we could get there; granted I don't think
fixing up ffs+softdep is the best way, and when we do get there
removing or quarantining softdep is definitely a reasonable thing to
do.

I'm just saying that while we have several choices that are *all*
problematic in one way or another, it's premature to be removing them
on the grounds that those problems exist.

-- 
David A. Holland
dholland%netbsd.org@localhost