From owner-freebsd-questions  Mon Jul 15 00:58:26 1996
Return-Path: owner-questions
Received: (from root@localhost)
          by freefall.freebsd.org (8.7.5/8.7.3) id AAA13336
          for questions-outgoing; Mon, 15 Jul 1996 00:58:26 -0700 (PDT)
Received: from root.com (implode.root.com [198.145.90.17])
          by freefall.freebsd.org (8.7.5/8.7.3) with ESMTP id AAA13329
          for <freebsd-questions@freebsd.org>; Mon, 15 Jul 1996 00:58:23 -0700 (PDT)
Received: from localhost (localhost [127.0.0.1]) by root.com (8.7.5/8.6.5) with SMTP id AAA02464; Mon, 15 Jul 1996 00:58:22 -0700 (PDT)
Message-Id: <199607150758.AAA02464@root.com>
X-Authentication-Warning: implode.root.com: Host localhost [127.0.0.1] didn't use HELO protocol
To: bill@twwells.com (T. William Wells)
cc: freebsd-questions@freebsd.org
Subject: Re: system hangs? after resetting rtq_reallyold 
In-reply-to: Your message of "15 Jul 1996 03:02:44 EDT."
             <4scqek$1kr@twwells.com> 
From: David Greenman <davidg@root.com>
Reply-To: davidg@root.com
Date: Mon, 15 Jul 1996 00:58:22 -0700
Sender: owner-questions@freebsd.org
X-Loop: FreeBSD.org
Precedence: bulk

>In article <199607150521.WAA02093@root.com>,
>David Greenman  <davidg@Root.COM> wrote:
>: options "NMBCLUSTERS=<n>"
>:
>:    ...to your kernel config file. Where <n> is some number in the area of 2000
>: and not more than 4000 (unless you know what you're doing).
>
>It's worth noting that 2.0 versions didn't need anywhere near this
>amount. When I upgraded to 2.1, it looked like I needed about
>twice what I'd previously needed. I've since doubled that and
>have been happy; I've set it to 4096.

   2.0 had a bug in it that caused it to use regular malloced buffers when it
couldn't get mbuf clusters. It would do this until it deleted most or all of
the kernel malloc area (and consume most of your memory in the process). The
end result in any case was significantly reduced performance. 2.0.5 and later
use mbuf clusters when it is reasonable to do so, and results in consistantly
high performance. This can be a bit more memory wasteful, however, than the
previous approach.
   It shouldn't be necessary to set NMBCLUSTERS that high on most "normal"
systems. It would take several hundred TCP connections to exhaust 2000 mbuf
clusters.

-DG

David Greenman
Core-team/Principal Architect, The FreeBSD Project