Message-ID: <002101bf68de$142f8430$5d01a8c0@p2350>
From: "emanuel stiebler" <emu@ecubics.com>
To: <pups@minnie.cs.adfa.oz.au>
References: <200001270142.RAA00523@moe.2bsd.com> <20000127145602.P53307@freebie.lemis.com> <20000127180406.A574@begemot.org> <20000127092641.A16017@rek.tjls.com>
Subject: Re: Sun release source code for Solaris 8
Date: Thu, 27 Jan 2000 08:49:19 -0700
Sender: owner-pups@minnie.cs.adfa.edu.au
Precedence: bulk

----- Original Message -----
From: Thor Lancelot Simon <tls@rek.tjls.com>
To: Joerg B. Micheel <joerg@begemot.org>
Cc: <pups@minnie.cs.adfa.oz.au>
Sent: Thursday, January 27, 2000 07:26
Subject: Re: Sun release source code for Solaris 8


> The dirty little secret of Linux and *BSD is that their ascendance has
> been tightly coupled to Sun's utter inability to build fast, cheap
> uniprocessor machines.  Any way you slice it, a single-processor top-of-
> the-line x86 box is just going to be a *lot* faster and cheaper than
> Sun's entry-level multiprocessor.  The great gamble they made was to
> turn their kernel into a highly-multithreaded thing of beauty -- but
> that *has* to cost some (even some small) amount of uniprocessor
> performance, and since they can't build cheap multiprocesors that are
> as fast as the high end of the commodity uniprocessor x86 boxes,
> for a lot of applications they lose.
>
> Even on a 2- or 4- processor machine, Solaris is demonstrably far
> faster than *BSD or Linux for many workloads.  But you can buy a
> single-processor x86 that's cheaper than Sun's 2- or 4- processor
> box now, which is why people run Linux or FreeBSD or NetBSD.  There
> is still a point at which the only way to get enough performance is
> to add more processors, and at that point Solaris still wins, and
> will for the forseeable (by me, at least) future.

Another thing to mention is also, that it is very easy to build your own
kernel, exctly for your needs in Linux or *BSD. (removing all
emulations/compatibility modes, ...) so you get a nice small/fast kernel
excactly for your type of machine & workload.

Don't think it's so easy on a sun.

cheers,
emanuel