From owner-freebsd-isp  Sun Nov 12 23:40:18 1995
Return-Path: owner-isp
Received: (from root@localhost)
          by freefall.freebsd.org (8.6.12/8.6.6) id XAA14114
          for isp-outgoing; Sun, 12 Nov 1995 23:40:18 -0800
Received: (from dyson@localhost)
          by freefall.freebsd.org (8.6.12/8.6.6) id XAA14060
          ; Sun, 12 Nov 1995 23:40:08 -0800
From: John Dyson <dyson>
Message-Id: <199511130740.XAA14060@freefall.freebsd.org>
Subject: Re: ISP state their FreeBSD concerns
To: jehamby@lightside.com (Jake Hamby)
Date: Sun, 12 Nov 1995 23:40:07 -0800 (PST)
Cc: mlh@poe.b2.org, dufault@hda.com, current@freebsd.org,
        freebsd-isp@freebsd.org
In-Reply-To: <Pine.BSF.3.91.951112212901.244A@localhost> from "Jake Hamby" at Nov 12, 95 09:31:32 pm
X-Mailer: ELM [version 2.4 PL24]
Content-Type: text
Content-Length: 1425      
Sender: owner-isp@freebsd.org
Precedence: bulk

> 
> Perhaps this is the reason behind the common observation that "Linux 
> freezes to a crawl when it starts swapping to disk."  I was told this was 
> one big advantage of FreeBSD.  Somehow, I'm starting to think that both 
> OS's have this problem, but perhaps this is less noticed under FreeBSD 
> due to:
> 
> 1) Different paging algorithms between FreeBSD and Linux (?)
> 2) More Linux users use IDE;  more FreeBSD users use SCSI.
> 
The paging algorithms on both stable versions of Linux and versions
through 1.3.20 (the last one that I tested) are much slower than
FreeBSD..  Also, sequential I/O is slower on EXT2FS.  The problem
that has been seen with FreeBSD with directory lookups is due to
several items, the biggest is the limited number of vnodes that
FreeBSD can cache.  I am actively looking at it, quantifying the
problem.  The benchmarks that Joe Greco has shown us actually
demonstrates a worst case clash with the name/vnode caching.  It *does*
represent a problem and if the files are opened in the inverse order
with my new async changes, it runs MUCH faster.  I am NOT belittling
the problem, just trying to explain that the problem is being worked
and studied.

The problem with going to sleep is probably due to the 30 second
sync interval and all of the modified vnodes (creation, modification,
usage) times and other things like that.  I have been looking at that
also.

John
dyson@freebsd.org