Please post...Bionet.biology.computational

ernest at lenti.med.umn.edu ernest at lenti.med.umn.edu
Fri Sep 27 01:39:50 EST 1991


> >A 20% difference in the speed of the CPU is nothing compared to getting
> >another nice screen.  In sum, buy as many workstations as you can afford.
> 
> I think this is nonsense.  Most of the time spent coding can be done as
> easily on an ascii terminal as on a "nice screen."  So buy the fanciest
> workstation you can afford, add a serial port card, and hook on as many 
> ascii terminals (or X-terminals, or Macs and PCs) as you like.  I have
> a bunch of old ascii terminals attached to my Personal Iris, and I
> almost never log onto the console -- but it's there when I need it.

I would tend to dramatically disagree with the latter, and agree with the
former.  Workstations are generally intended to be just that--a wonderful
*personal* facility.  While they are virtually all multi-user and multi-
tasking, they were not intended to be a mini.  A server, with, for example,
a better data bus and all those techie things that make up a server, yes; 
but then you have lost the price-point of a workstation, and, by and large, 
most of it's functionality as a workstation [we make ours do double duty].  
Most of the programmer's I know prefer at least the functionality of an 
X-terminal; but as far as general use, one has to remember that the 
computation is being done on the client [yes, that means, on what most of 
think of as the server].  If one is not careful what you hang those off of, 
those wonderful X-terminals will grind a workstation to a halt. These work 
best hung off a machine that was intended for that type of use--a multi-
processor Sequent or SGI, or a Sun server [or Cray, for that matter].

> Bang-per-buck is still the
> right criterion, however, just as long as we realize that (FLOPS != bang).

Most computations in molecular biology-related work and similarity searching 
are not, however, floating point.  

>    ...Sun is behind in pure CPU floating point speed
> 
> Not just fp!  Sun is way behind, period.  I recently posted to
> comp.benchmarks the results of running sim, a program which computes
> the top N best local alignments in two dna or protein sequences, on a
> variety of machines.  Sun was at the bottom.  HP was at the top,
> besting Cray.  This is integer, scalar code, not fp at all.

Cray's were never designed to do integer, scalar code.  If it is not
fp-intensive, or the code does not vectorize well, and likely if it is
written in C rather than that wonderful language of supercomputers, Fortran,
there is really no reason to include the Cray in a test suite.  Like the
Connection Machine, they were never meant to be a general-purpose computer.

In the Chip Wars, whoever is ahead today is of interest mostly to the
sales rep for the company.  Most chemists I know are dreaming of the IBM
box to put in the corner for a "cooker," and use the SGI's for display.
What happens when the Sun comes in with the TI superscalar chip?  What does
someone who bought super-bang last week do then?

I tend to think in more practical terms, by and large.  What machines will
do what I want in a reasonable amount of time, what is the software 
availability, what is the cost of that software [remember, the same program
written for different OS and windowing systems is generally priced depending
on how many units they expect to sell], what is the level of system
administration you want to do, how easy will it be to incorporate into an
environment of other manufacturers, and how many different operating systems 
do you want to support?  We have an environment where practically everything
imaginable is available; what gets used?  Suns, far and away; X-terminals,
generally on real compute-servers; Irises for most graphics-related work.  
Cost/seat: Sun, 1X; X term, 0.5X; Iris, ~1.5X or 4X [depending on whether you 
are talking about an Indigo or a PI]; full-blown Irises--well, inquire, but 
be sitting down when you do...

These are my opinions; I have no relationship with Sun, SGI, IBM, 
Cray, Sequent, Thinking Machines, HP, TI, any X-term manufacturer, or any 
other corporate entity that you might imagine I was supporting.  But I do
have opinions about all of them, and more!!

Ernie Retzel
Dept. of Microbiology
University of Minnesota

ernest at lenti.med.umn.edu
--
Domain: curtiss at umiacs.umd.edu		     Phillip Curtiss
  UUCP:	uunet!mimsy!curtiss		UMIACS - Univ. of Maryland
 Phone:	+1-301-405-6710			  College Park, Md 20742



More information about the Comp-bio mailing list