From owner-freebsd-hackers  Mon Nov  4 06:58:38 1996
Return-Path: owner-hackers
Received: (from root@localhost)
          by freefall.freebsd.org (8.7.5/8.7.3) id GAA22808
          for hackers-outgoing; Mon, 4 Nov 1996 06:58:38 -0800 (PST)
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 GAA22793
          for <freebsd-hackers@freefall.freebsd.org>; Mon, 4 Nov 1996 06:58:33 -0800 (PST)
Received: (from jgreco@localhost) by brasil.moneng.mei.com (8.7.Beta.1/8.7.Beta.1) id IAA05007; Mon, 4 Nov 1996 08:57:06 -0600
From: Joe Greco <jgreco@brasil.moneng.mei.com>
Message-Id: <199611041457.IAA05007@brasil.moneng.mei.com>
Subject: Re: Another data point in the daily panics...
To: gpalmer@freebsd.org (Gary Palmer)
Date: Mon, 4 Nov 1996 08:57:05 -0600 (CST)
Cc: ponds!rivers@dg-rtp.dg.com, greg@uswest.net,
        freebsd-hackers@freefall.freebsd.org
In-Reply-To: <4048.847003445@orion.webspan.net> from "Gary Palmer" at Nov 3, 96 01:44:05 am
X-Mailer: ELM [version 2.4 PL24]
Content-Type: text
Sender: owner-hackers@freebsd.org
X-Loop: FreeBSD.org
Precedence: bulk

> Thomas David Rivers wrote in message ID
> <199611011143.GAA11189@lakes.water.net>:
> > 	2) Does INN & CCD hit the file system as hard as a CNEWS expire?
> > 	   (I wouldn't think so with an mmap'd active file.  But then
> > 	    you probably shouldn't mmap the active file because of
> > 	    problems there.)
> 
> I would be surprised if the drive loading was different. The only
> thing that may change is the accessing of the drive where the
> history/active file(s) are kept, and that shouldn't affect the CCD's
> spool... all I know is when expireover/fastrm hits my FS, I can kiss
> the performance goodbye.

You are not kissing your performance totally goodbye if you have your
news spool on multiple filesystems...  but I assume you have them all
striped into one mongo filesystem.

Personally I wanted to see fastrm go as fast as possible, so I have six
spool filesystems each built out of two Hawk drives CCD'd together...

I run a special "expirerm" script that knows which hierarchies are on
which disks, and then goes and invokes "fastrm" in parallel.  This has
the side effect of taking the machine to its knees for a few minutes,
as I/O becomes saturated, but it takes 1/6 as long as doing it on a
larger unified CCD spool..  hey, if you are going to take a performance
hit, take a REAL performace hit but minimize the amount of time.

(And no, I do not want to hear about the 'ease of administration' issue
of a single spool from ANYBODY :-)  )

newspump.sol.net, #25 on the Freenix list this month.  :-)  Pure FreeBSD.

... JG