
					Nickolai Zeldovich, 1996

This program (programs actually) let you do a neuron net simulation on
your Apollo node(s).

Here is its basic structure:

      DLD <----- GLD <----> LLDs

The GLD is the "hub" module.
The DLD is the display module, requires an Apollo node (uses GPR).
The LLD is the calculational module, it forks N of itself for each
"animal", as specified in config.h

All of these can run on different nodes.

A DN10000 or a non-Apollo node should be used for GLD's and LLD's, because
slower Apollo nodes can't do it very well (DN3500, DN4500).

Here is what the files' extensions are:

.c     - C source code
.m     - M68k compiled binary
.a     - A88k (DN10k) compiled binary
.linux - Linux, ELF

The gld.a, gld.m, lld.a, lld.m are coff-type executables.
The dld.m is a obj-type executable (SR9).

You have to set the type using /com/obty otherwise it won't run.

How to start this:

(a) First, start a gld.a or gld.m.

(b) Then, give it a random number to start with.

(c) Now start your DLD by typing
% dld.m gldhost

(d) Then give it the number of nodes which will run LLD's (you can split
the load by putting half of the forked LLD's on one node and half on
another, or on 3 or more nodes).

(e) Then, start LLDs by doing:
% lld.a gldhost
or
% lld.m gldhost
on every one of the nodes specified above in (d).

You should be running by now.

If you want to compile the GLD/LLD on any other platform, use a different
config.h file (config.h.2) if the normal one doesn't work. For example, I
found out that Linux doesn't like #define's, and prefers const's there.
Apollo does the opposite.

If you have any questions, mail them to me at kolya@zepa.net.
