Path: sparky!uunet!amdahl!rtech!sgiblab!spool.mu.edu!hri.com!
noc.near.net!ns.draper.com!news.draper.com!surname!ibe1109
From: ibe...@draper.com (Ira Ekhaus)
Newsgroups: comp.os.linux
Subject: Is Linux a viable OS for a stable multiuser unix system, 
not just a hobbiest's Unix box?
Message-ID: <IBE1109.93Feb9142158@etbsun1.draper.com>
Date: 9 Feb 93 19:21:58 GMT
Sender: nn...@draper.com (NNTP Master)
Organization: Draper Laboratory, Inc., Cambridge, MA
Lines: 35
Nntp-Posting-Host: etbsun1.draper.com


Hi,

I'm interested in setting up a Unix box for a few researchers.  It
would be doing some fairly light computing (by scientific computing
standards).

I'm trying to determine the tradeoffs between

The lower cost of  *86 boxes compared to  sparc's. 
vs
the delays and distractions that an unstable operating system would
provide.

As the prospective system gets bigger the relative cost of the CPU
becomes less of an issue (peripherals costing the same on both systems
)  and I believe the SPARC workstation wins out.

But for a low capital interim solution Linux might work because
	1) the code I'd write would be unix generic
	2) the 486 box is already there and 
	3) at the very least the 486 box would make
	a good Xterm (using linux of course).

Has anyone else considered these tradeoffs?

thanks,

Ira
ekh...@draper.com
Phone: 617 258 1109
Draper Laboratories m.s. 7c
555 Technology Square
Cambridge , MA 
02139

Path: sparky!uunet!newsgate.watson.ibm.com!news.ans.net!
howland.reston.ans.net!usc!cs.utexas.edu!qt.cs.utexas.edu!yale.edu!
ira.uka.de!rz.uni-karlsruhe.de!fg30.rz.uni-karlsruhe.de!ig25
From: ig...@fg30.rz.uni-karlsruhe.de (Thomas Koenig)
Newsgroups: comp.os.linux
Subject: Re: Is Linux a viable OS for a stable multiuser unix system, 
not just a hobbiest's Unix box?
Date: 9 Feb 93 21:48:48 GMT
Organization: University of Karlsruhe, Germany
Lines: 53
Message-ID: <ig25.729294528@fg30>
References: <IBE1109.93Feb9142158@etbsun1.draper.com>
NNTP-Posting-Host: fg30.rz.uni-karlsruhe.de
Mime-Version: 1.0
Content-Type: text/plain; charset=iso-8859-1
Content-Transfer-Encoding: 8bit

ibe...@draper.com (Ira Ekhaus) writes:

>I'm interested in setting up a Unix box for a few researchers.  It
>would be doing some fairly light computing (by scientific computing
>standards).

What is light?  How many megs of ram do you expect to need, and how
many megs of data do you expect to generate?

>I'm trying to determine the tradeoffs between

>The lower cost of  *86 boxes compared to  sparc's. 
>vs
>the delays and distractions that an unstable operating system would
>provide.

Depends.  I'd say (personal experience) that Linux is stable enough for
lightweight scientific stuff.  Where it misses out is on the
availability of commercial software like the NAG libs, Maple, ...  etc.
Of course, if a copy of Numerical Recipies is enough for your numerical
needs, you're fine.  F2c, while being far from great (especially because
of difficult source - level debugging) does a reasonable job at
compiling even fairly largish portions of FORTRAN code.  Oh, and just
wait until I've uploaded fudgit to tsx-11 (should happen in the next few
days :-)

Always keeping up to date is work, sure, but it's not too bad if
your box happens to be connected to the Internet.

>As the prospective system gets bigger the relative cost of the CPU
>becomes less of an issue (peripherals costing the same on both systems
>)  and I believe the SPARC workstation wins out.

Depends...  if your applications require > 16 Meg real RAM and SCSI
drivers, go for the SPARC.  If you are going to add gigabyte storage,
use the SPARC as woll.  If your plans are not as ambitious, I'd say
Linux is a viable alternative.

>But for a low capital interim solution Linux might work because
>	1) the code I'd write would be unix generic
>	2) the 486 box is already there and 
>	3) at the very least the 486 box would make
>	a good Xterm (using linux of course).

Those three arguments certainly count, if nothing else does :-)

>Has anyone else considered these tradeoffs?

Yep... I have :-)
-- 
Thomas Koenig, ig...@rz.uni-karlsruhe.de, ig25@dkauni2.bitnet
The joy of engineering is to find a straight line on a double
logarithmic diagram.

Newsgroups: comp.os.linux
Path: sparky!uunet!news.claremont.edu!elroy.jpl.nasa.gov!swrinde!
zaphod.mps.ohio-state.edu!howland.reston.ans.net!atlantis.psu.edu!
wintermute.phys.psu.edu!ra!tantalus.nrl.navy.mil!eric
From: er...@tantalus.nrl.navy.mil (Eric Youngdale)
Subject: Re: Is Linux a viable OS for a stable multiuser unix system, 
not just a hobbiest's Unix box?
Message-ID: <C28zzu.1q7@ra.nrl.navy.mil>
Sender: use...@ra.nrl.navy.mil
Organization: Naval Research Laboratory
References: <IBE1109.93Feb9142158@etbsun1.draper.com> <ig25.729294528@fg30>
Date: Wed, 10 Feb 1993 19:40:41 GMT
Lines: 35

In article <ig25.729294528@fg30> ig...@fg30.rz.uni-karlsruhe.de (Thomas Koenig) 
writes:
>Depends...  if your applications require > 16 Meg real RAM and SCSI
>drivers, go for the SPARC.  If you are going to add gigabyte storage,
>use the SPARC as woll.  If your plans are not as ambitious, I'd say
>Linux is a viable alternative.

	I feel that this should probably be qualified a bit.  First of all,
the current scsi drivers are quite capable with dealing with machines that have
more than 16Mb.  There is a slight overhead as the disk blocks are copied to
memory locations < 16Mb, but other than a barely noticable performance hit, it
will work just fine.

	If I were going to try and set up a machine for number crunching, I
would probably go with either local bus or EISA for the disk controller.  We
have two machines here that we use for moderate number crunching with SVr4, and
both have EISA motherboards.  At the time we bought the machines the EISA disk
controllers were still quite expensive, so we went with ISA controllers - it
may not be the best, but we have an easy upgrade path available to us if the
need arises.

>>Has anyone else considered these tradeoffs?
>
>Yep... I have :-)

	So have I.  When we made the choice (~2-3 years ago), linux was not
available, so we went with SVr4.  At the time, there was a real price premium
for 483-33 machines, and I think that a loaded system with SVr4 went for around
10K$.  A stripped EISA machine went for around 5K$.  We could have gotten an
entry level Sparc machine for around 20K$ (stripped it would have been around
10K$), but as I recall the price/performance ratios were about the same for the
two machines, in terms of number crunching.

-Eric

--