From owner-freebsd-commit  Mon Nov  6 08:04:11 1995
Return-Path: owner-commit
Received: (from root@localhost)
          by freefall.freebsd.org (8.6.12/8.6.6) id IAA29788
          for freebsd-commit-outgoing; Mon, 6 Nov 1995 08:04:11 -0800
Received: (from root@localhost)
          by freefall.freebsd.org (8.6.12/8.6.6) id IAA29773
          for cvs-all-outgoing; Mon, 6 Nov 1995 08:04:03 -0800
Received: (from root@localhost)
          by freefall.freebsd.org (8.6.12/8.6.6) id IAA29759
          for cvs-sys-outgoing; Mon, 6 Nov 1995 08:04:00 -0800
Received: (from dyson@localhost)
          by freefall.freebsd.org (8.6.12/8.6.6) id IAA29740
          ; Mon, 6 Nov 1995 08:03:40 -0800
From: John Dyson <dyson>
Message-Id: <199511061603.IAA29740@freefall.freebsd.org>
Subject: Re: cvs commit: src/sys/ufs/ufs ufs_lookup.c
To: bde@zeta.org.au (Bruce Evans)
Date: Mon, 6 Nov 1995 08:03:38 -0800 (PST)
Cc: j@uriah.heep.sax.de, CVS-commiters@freefall.freebsd.org,
        cvs-sys@freefall.freebsd.org
In-Reply-To: <199511061550.CAA22468@godzilla.zeta.org.au> from "Bruce Evans" at Nov 7, 95 02:50:50 am
X-Mailer: ELM [version 2.4 PL24]
Content-Type: text
Content-Length: 998       
Sender: owner-commit@FreeBSD.org
Precedence: bulk

> 
> >>   Modified:    sys/ufs/ufs  ufs_lookup.c
> >>   Log:
> >>   This commit causes UFS to perform at Linux EXT2FS metadata rates.
> >>   ...  Just use the -o async flag on the
> >>   UFS mount.  Good luck...
> 
> >Great!
> 
> >I will stress-test this on my /tmp partition!
> 
> I think the MNT_ASYNC check in ffs_update() goes too far.  It stops
> timestamps (and perhaps more important things) from being written
> immediately even for sync() and fsync().
> 
> Bruce
> 
The sync system call (and the update daemone) works correctly
because the MNT_ASYNC flag is
temporarily cleared.  You argument for fsync is more valid though, and
will be corrected very soon.

Actually, I am considering supporting a couple of levels of asynchness,
a "safe" one that just helps speed things up with little risk, and
a really fast "no holds barred" one like Linux.  Perhaps the "safe" one
will do meta-data updates every second or so, and the unsafe one will work
like it does now??

John
dyson@freebsd.org