Subject: Re: Summer of code ideas
To: Johan A.van Zanten <johan@giantfoo.org>
From: Thor Lancelot Simon <tls@rek.tjls.com>
List: netbsd-users
Date: 03/24/2007 11:49:00
On Fri, Mar 23, 2007 at 09:35:26PM -0500, Johan A.van Zanten wrote:
> 
> Bill Stouder-Studenmund <wrstuden@netbsd.org> wrote:
> > Thank you. However we were wanting to compare soft updates with a 
> > journaled ffs, not LFS.
> 
>  I'm sorry, i'm a little confused now.  What you previously said was:
> 
> "Do you have any recent references? Specifically NetBSD-based ones?"
> 
>  Is there a journaled FFS in NetBSD?  If not, how was i to compare this
> all on NetBSD?  Or were you suggesting that we compare NetBSD FFS + soft
> updates against a journaled file system on another OS like Solaris or AIX?

To be blunt, if you don't understand the difference between a filesystem
like FFS with transaction journaling added and a log-structured filesystem,
you're probably not competent to design tests that expose any meaningful
metric of performance between two kinds of filesystems.

Since NetBSD does not have an implementation of journaling for FFS, and
Solaris does not have an implementation of soft updates, to actually test
this right you would need to do at least three benchmarks with the same
hardware, and _hope_ that you got numbers that were comparable -- which the
first comparison might show you you had not.  You'd need to test Solaris FFS
and NetBSD FFS without journaling or soft updates, then, for the same
workload on the same hardware, Solaris FFS with journaling and NetBSD FFS
with soft updates.

The problem of course is that you might well find enough disparity in
performance in the first step that you could not extract a meaningful
number.  To actually be guaranteed useful results, you would have to have
FFS journaling and soft update implementations of good quality that you
could run in the _same_ kernel, on the same hardware.  The only paper I
know of that claims to present such results is Seltzer's, but there has
been a persistent difficulty actually obtaining the code Seltzer uses to
run her benchmarks, going all the way back to her LFS work.  This is why
I am highly skeptical of the literature in this area.

Thor