Return to BSD News archive
Path: euryale.cc.adfa.oz.au!newshost.carno.net.au!harbinger.cc.monash.edu.au!nntp.coast.net!trellis.wwnet.com!ddsw1!news.mcs.net!not-for-mail
From: tundra@MCS.COM (Tim Daneliuk)
Newsgroups: comp.unix.bsd.freebsd.misc
Subject: Re: FreeBSD and Linux
Date: 4 Oct 1996 09:05:50 -0500
Organization: TundraWare
Lines: 47
Message-ID: <5335ju$sv5@Mercury.mcs.com>
References: <3246f8e0.1466924@news.telepac.pt> <324924E5.49B6@usoft.nl> <324AC49E.1CD3@pressconnect.com> <32507B89.1CFBAE39@freebsd.org>
Reply-To: tundra@tundraware.com
NNTP-Posting-Host: mercury.mcs.com
In article <32507B89.1CFBAE39@freebsd.org>,
Jordan K. Hubbard <jkh@FreeBSD.org> wrote:
>Mischka Hughes wrote:
>>
>> How about robustness - I need a server / OS combination that is totally
>> stable. It must never go down. I am prepared to pay the price of a
>> little slowness for robustness.
>>
>> Now what do you recommend, FreeBSD or Linux or even Linux-FT.
>
>None of the above, nor NT or any of the commercial alternatives.
>
>There is no such thing as a server / OS combination that is totally
>stable and never goes down - such things are currently too far beyond
>our level of technological sophistication to build. All the OS groups,
>free and commercial, have written software which is more or less stable
>in some ways than the others, and deciding which one is right for you is
>a decision which should be driven far more by the assessment of
>available personnel than any marketing organization's claims of "total
>stability" - anyone who tells you they can offer that probably also has
>very reasonably priced lunar vacations to sell ("every room offering
>splendid views of the Tyco crater!") if you know to ask about them.
>
Actually, there is a machine/OS combination that comes very close to
the goal of 100% uptime and that is a Tandem running the Non-Stop Kernel.
To all intents and purposes, this machine will "never" go down because
it is fault tolerant in both hardware *and* software.
>Your own best bet is probably to spend as much money as you can on a
>commercial package. It won't be any more reliable, but you'll have at
>least paid for the privilege of yelling at someone over the phone about
>it.
>--
A very reasonable strategy here is to use the 'n+1' strategy we used
when I worked on the United Airline Apollo reservations system. This
system was only down about 5-10 minutes per *year* typically. The
idea is to always have one more machine/OS/applications software image
available than was actually needed in worst case load. That way, if
you had a failure, you always had a fallback environment. When not
needed on-line, the 'n+1'th machine was a test and development
environment.
--
------------------------------------------------------------------------------
Tim Daneliuk / tundra@tundraware.com
Voicemail/FAX 847.827.1706