From owner-freebsd-stable  Sat May 10 22:39:45 1997
Return-Path: <owner-stable>
Received: (from root@localhost)
          by hub.freebsd.org (8.8.5/8.8.5) id WAA13363
          for stable-outgoing; Sat, 10 May 1997 22:39:45 -0700 (PDT)
Received: from scanner.worldgate.com (scanner.worldgate.com [198.161.84.3])
          by hub.freebsd.org (8.8.5/8.8.5) with ESMTP id WAA13357;
          Sat, 10 May 1997 22:39:41 -0700 (PDT)
Received: from znep.com (uucp@localhost)
	by scanner.worldgate.com (8.8.5/8.8.5) with UUCP id XAA20138;
	Sat, 10 May 1997 23:39:34 -0600 (MDT)
Received: from localhost (marcs@localhost) by alive.znep.com (8.7.5/8.7.3) with SMTP id XAA04640; Sat, 10 May 1997 23:36:49 -0600 (MDT)
Date: Sat, 10 May 1997 23:36:48 -0600 (MDT)
From: Marc Slemko <marcs@znep.com>
To: Gary Palmer <gpalmer@FreeBSD.ORG>
cc: freebsd-stable@FreeBSD.ORG
Subject: Re: INN and mmap() 
In-Reply-To: <4032.863322785@orion.webspan.net>
Message-ID: <Pine.BSF.3.95.970510232912.220F-100000@alive.znep.com>
MIME-Version: 1.0
Content-Type: TEXT/PLAIN; charset=US-ASCII
Sender: owner-stable@FreeBSD.ORG
X-Loop: FreeBSD.org
Precedence: bulk

On Sat, 10 May 1997, Gary Palmer wrote:

> Chad R. Larson wrote in message ID
> <9705060107.AA18210@chad.anasazi.com>:
> >    2) Is the 2.1 mmap() broken, or is there just a misunderstanding
> >       between the implementation and INN's expectations?
> 
> Yes to both. 
> 
> >    3) Has mmap() been changed in 2.2 or 3.0?  If so, are the changes
> >       slated to be rolled into the RELENG_2_1_0 tree?
> 
> RELENG_2_1_0 is a dead tree for stuff like this (and I believe the
> changes were too extensive too). Believe me, from experience,
> RELENG_2_2 works *MUCH* better for heavily hit servers like news
> servers. news.webspan.net has been running from the RELENG_2_2 branch
> for well over 8 months without any kernel problems, and with working
> mmap() makes things a bit better (some other local tweaks help too :) )

We upgraded a news server from INN 1.4unoff4 + patches running on
RELENG_2_1_0 to 1.5.1 running on RELENG_2_2 and I was amazed by how much
less memory the new server took.  innd took a third of the memory that it
did before. 

Not sure how much of that is due to 1.5.1 vs. 1.4unoff4, but a lot is due
to the much better malloc() in 2.2.

Only problem we have with that server is that once in a while the
profiling clock (aka. "alternate system clock" by systat -v) dies and
needs a poke to restart.  Isn't a huge deal since I just restart it from
the initialization code of a lkm, but it cna be annoying.