From owner-freebsd-hackers  Mon Jun 19 11:32:09 1995
Return-Path: hackers-owner
Received: (from majordom@localhost)
          by freefall.cdrom.com (8.6.10/8.6.6) id LAA03631
          for hackers-outgoing; Mon, 19 Jun 1995 11:32:09 -0700
Received: from Root.COM (implode.Root.COM [198.145.90.1])
          by freefall.cdrom.com (8.6.10/8.6.6) with ESMTP id LAA03623
          for <freebsd-hackers@freebsd.org>; Mon, 19 Jun 1995 11:32:06 -0700
Received: (dyson@localhost) by Root.COM (8.6.11/8.6.5) id LAA05790; Mon, 19 Jun 1995 11:31:57 -0700
Date: Mon, 19 Jun 1995 11:31:57 -0700
From: John Dyson <dyson@Root.COM>
Message-Id: <199506191831.LAA05790@Root.COM>
To: freebsd-hackers@freebsd.org, graichen@sirius.physik.fu-berlin.de
Subject: Re: freebsd and memory
Sender: hackers-owner@freebsd.org
Precedence: bulk

FreeBSD uses both a buffer cache and a VM cache.  It is probably using
more than 3MB for its buffer cache on a 16MB system.  The kernel data
structures on FreeBSD are not small either.  Note also that FreeBSD
does NOT defer updates to filesystem metadata.  Linux does -- so it
is more possible to have stale metadata on a Linux disk.  FreeBSD
also starts filesystem writes almost immediately when a file is written
to (actually when a cluster is completed.)  I think that Linux waits
until a sync occurs (or the "sync" daemon runs.)

I have been pondering the relative merits of different policies for
filesystem I/O.  Some experiments that I have run show that if the
data needs to be non-volatile on the disk that the system is faster
to do the I/O like FreeBSD does.  If the data is volatile, it appears
to be faster to do the I/O like Linux does.  (These are just heuristic
type off-the-cuff observations.)

John
dyson@root.com