From: sorrow@oak.circa.ufl.edu
Subject: Re: TECH: Synchronized communications between various machines
Date: Monday, Apr 6 01:18:00 1992
Organization: The CIRCA Underground


|>Treat your system of interlinked machine as one system and have the 
|>faster machines do the number crunching and the slower machines all the 
|>other processes,  then feed this information onto your buses with an
|>indectification for each different type of data.  Like packet 
|>I think.  That way you should get a FASTER system as a whole than a 
|>slower system.
|>
|>Each machine would be fed it's share of the programs when it 'logs on'
|>so the more machines there are the more power you have.
|>
|>Each machine would have a base program,  every machine would have 
|>this, to run the world and the communications then other programs to
|>be run as subroutines  (turbo c like) when they log on.

I thought of this method, however it seemed to me when I did some
offhand calculations that the speed of the transfer would be SLOWER
than if the client machine had just done the math itself!  For example,
would it be faster to send a 4x4 matrix of 32-bits ( 512-bits ) for
the rotation and translation, for 1024 bits, plus about ten bytes of
other data.  This works out to about 1100 bits.  If I want to support
a serial line, for two machines for now, this would be about 1100 bits
for every update.  Thus each update could be transfered at about 1/8th
or so of a second.

So, I guess the question still is:  will calculations, on the average,
for the 4 x 4 matrices be faster than about 120msec?  I think so.  So
that leaves us at our original problem -- having faster machines
synch to slower machines without having them bog down.

Any suggestions or corrections are welcome.