From owner-freebsd-isp  Mon Aug 26 10:46:55 1996
Return-Path: owner-isp
Received: (from root@localhost)
          by freefall.freebsd.org (8.7.5/8.7.3) id KAA26988
          for isp-outgoing; Mon, 26 Aug 1996 10:46:55 -0700 (PDT)
Received: from brasil.moneng.mei.com (brasil.moneng.mei.com [151.186.109.160])
          by freefall.freebsd.org (8.7.5/8.7.3) with ESMTP id KAA26976
          for <freebsd-isp@freebsd.org>; Mon, 26 Aug 1996 10:46:47 -0700 (PDT)
Received: (from jgreco@localhost) by brasil.moneng.mei.com (8.7.Beta.1/8.7.Beta.1) id MAA00687; Mon, 26 Aug 1996 12:45:04 -0500
From: Joe Greco <jgreco@brasil.moneng.mei.com>
Message-Id: <199608261745.MAA00687@brasil.moneng.mei.com>
Subject: Re: Anyone using ccd (FreeBSD disk striper) for news
To: michaelv@MindBender.serv.net (Michael L. VanLoon)
Date: Mon, 26 Aug 1996 12:45:04 -0500 (CDT)
Cc: jgreco@brasil.moneng.mei.com, michael@memra.com, freebsd-isp@freebsd.org,
        craigs@os.com
In-Reply-To: <199608261720.KAA16235@MindBender.serv.net> from "Michael L. VanLoon" at Aug 26, 96 10:20:02 am
X-Mailer: ELM [version 2.4 PL24]
Content-Type: text
Sender: owner-isp@freebsd.org
X-Loop: FreeBSD.org
Precedence: bulk

> >> And did you need to fo that tweak to UFS meta data updates that makes it
> >> more like Linux's ext2fs?
> 
> >I've played with metadata updates before (disabled ATIME and MTIME updates)
> >but I don't want to do it on a production system until there is an
> >officially sanctioned method to do so.
> 
> On the other hand, I've had two NetBSD machines running with all
> filesystems in async mode for several months, now.  One of these
> machines is getting beat since I'm running lots of benchmarks on it.
> The X benchmarks, in particular, seem to kill it quite regularly (must
> be bugs in the 3.1.2E beta server), forcing me to hit reset since X
> gets totally hung, and it's not networked to anything.
> 
> I have yet to lose a single file, that I know of, running my drives in
> async mode.  In fact, when it comes back up, I look to see how much of
> the data it finished collecting in the benchmark file it was writing
> before I reset it.  The file is always there.
> 
> Of course, your results may vary...

But async mode will not buy you much in this case...!  Async is simply
delaying the inevitable, which is already delayed in FreeBSD until the next
30-second sync.  I do not want to write metadata at _all_ if all it is is 
an ATIME update...  if I have to write it, now OR later, I am losing
bandwidth and transactions per second that I would rather spend reading
other data.

Now of course if you are doing OTHER things, async is a real win.

... JG