🚀 go-pugleaf

In article <4065@bcstec.ca.boeing.com> silverm@bcstec.ca.boeing.com (Jeff Silverman) writes:
>Gavin.Flower@comp.vuw.ac.nz (Gavin Flower) writes:
[...]
>>Performance will obviously be inferior to a well designed single host
>>GUI system like an Acorn Archimedes running Risc-OS, but "X-windows"
>>is designed to be run over a network.
>
>Gavin, 
>	Although there is some truth in what you say, my experience tells me
>that it ain't necessarily so.  No doubt there is overhead in formatting the
>packets, sending them across the net, decoding them on the other side, etc.
>However, the X-server relieves the X-client (the X-client is the program, but
>in this instance I use "X-client" to mean the computer system on which the
>program is running) of certain chores, such as manipulating bit maps, etc.
>Thus, you have 2 processors running in parallel, more or less.  Also, the 
>X-server can be implemented using a speciallized processor, such as one of the 
>TI34xxx  chips, while the client is implemented using a general processor, 
>such as a 68xxx, 80x86, or VAX etc.
>	My experience running the X demonstration ICO on a VAXstation 3100
>with VMS 5.4-3 vs. on a VXT 2000 connected to that same VAXstation is that
>ICO runs faster on the VXT.  That is the extent of my testing.  You might be
>right in the general case, but the answer isn't that simple.


Very true - the paragraph quoted from my may have seemed anti-network
and anti "X", but it wasn't meant to be.

I believe that the most cost effective way of giving a large number of
people lots of computing resources is to have a relatively small
number of high powered servers selected to be the best for the range
of software required and a larger number of workstations of various
sorts.  In this situation, it is unlikely that everybody will need a
really high end grsphics workstation, and some could probably be
serviced adequately with a character only terminal.  

There is some software that will spread the load over however many
hosts that are available.

Obviously, a single box can deliver more effective processing power,
as a percentage of what is theoretically, than a network.  However, a
network can allow the several boxes to cooperate, to deliver more
effective processing power than any single box, also this can be
tapped by anyone (who is authorised) on the network - they don't have
to be physically in front of a particular box.

How significant the performance degredation is, depends on the ratio
of the overhead involved to the gains, "obviously".

On a side note, I've been surprised at the slowness of window
refreshing on some machines with a high mips rating, compared to a
machine with a much lower one.  I could give brand names, but I'm not
trying to start a flame war.  Besides, care has to be taken to allow
for the different processing loads that the machines may have had,
that was not visible.  Though in the case I'm thinking of, this should
not have made as much diffrence.

The subject of objective benchmarking of distributed systems, has
come up, but I haven't seen much comment about it here - even about
the principles. 


Gavin
-- 
Tomorrow's Society                              | flower_g@kosmos.wcc.govt.nz
    Depends on                                   ----------------------------
Today's Children  - help us raise them properly! 
*******  These comments have no known correlation with dept. policy!  *******