From owner-freebsd-hackers  Sun May 26 10:10:59 1996
Return-Path: owner-hackers
Received: (from root@localhost)
          by freefall.freebsd.org (8.7.5/8.7.3) id KAA04758
          for hackers-outgoing; Sun, 26 May 1996 10:10:59 -0700 (PDT)
Received: from dyson.iquest.net (dyson.iquest.net [198.70.144.127])
          by freefall.freebsd.org (8.7.5/8.7.3) with ESMTP id KAA04753
          for <hackers@freebsd.org>; Sun, 26 May 1996 10:10:53 -0700 (PDT)
Received: (from root@localhost) by dyson.iquest.net (8.7.5/8.6.9) id MAA00878; Sun, 26 May 1996 12:09:46 -0500 (EST)
From: "John S. Dyson" <toor@dyson.iquest.net>
Message-Id: <199605261709.MAA00878@dyson.iquest.net>
Subject: Re: VM in FreeBSD and 4.4BSD book
To: T0682740@pmail.tepco.co.jp (Motonori Shindou)
Date: Sun, 26 May 1996 12:09:46 -0500 (EST)
Cc: hackers@freebsd.org
In-Reply-To: <31A889F0.6553.001@pmail.tepco.co.jp> from "Motonori Shindou" at May 27, 96 01:42:24 am
X-Mailer: ELM [version 2.4 PL24 ME8]
MIME-Version: 1.0
Content-Type: text/plain; charset=US-ASCII
Content-Transfer-Encoding: 7bit
Sender: owner-hackers@freebsd.org
X-Loop: FreeBSD.org
Precedence: bulk

> Hello.
> 
> Finally, I obtained "4.4BSD Book" and started reading it. In Section
> 5.5 of this book (p.145), it says:
> 
>   This behavior is an unwelcome side effect of the separate virtual
>   memory and filesystem caches; it would be eliminated if the two
>   caches were integrated.
> 
> Is this just what happned in FreeBSD 2.0.5 as described in its
> REL-NOTES?
> 
>   Merged VM-File Buffer Cache
>   ---------------------------
>   A merged VM/buffer cache design greatly enhances overall system
>   performance and makes it possible to do a number of more optimal
>   memory allocation strategies that were not possible before.
> 
Yes, actually we had several versions of prototype merged VM buffer cache
code since pre-1.1.5 days.  The version that we ended up using appeared
to be the cleanest way to do it at the time, by limiting most of the
changes in the system to vfs_bio.  Many of the changes to the filesystem
code are bugfixes due to misuse of the subroutine "vnode_pager_uncache."

The FreeBSD buffer cache code is also designed to limit pressure on pages in
processes.  It also limits the number of dirty pages from the buffer cache
to a reasonable size.  There are more things that we can do to improve the
managing buffer cache memory policy, but it hasn't been necessary so far.

Warning, even though a couple of FreeBSD people reviewed the VM section of
the 4.4BSD book, the VM system on FreeBSD has changed substantially.  The
book generally applies, but there have been many fixes and rewrites relative
to the original port of the VM code.  There is still more to do though!!!

John
dyson@freebsd.org