source: branches/OKBJavaConnector/ECJClient/src/ec/exchange/2-Island-Server/README @ 11194

[If you're looking for a simple example of doing an island
model, don't look here -- look instead in 3-Island.]

This model is a special example of how to create a server
that's not actually a client program -- the process
SOLELY exists to be a server.  Why would you want to do this?
Well, we were running on a beowulf cluster maintained by a 
company that wouldn't allow us to know beforehand exactly
what the IP address was of *any* node.  (They were using a
queueing system, argh).  We got around this by putting our
clients out on the cluster, but having a dedicated machine
with a known IP address as the server.  The company permitted
this because the server doesn't have almost any network traffic
going to it -- it's just for synchronization.

So in this example we'll have two islands, Minoa and Crete,
using minoa.params and crete.params, plus a separate server 
(in server.params).  You need to launch the server differently
than you'd normally launch an ECJ process.  What you do here
is launch the server as:

java ec.exchange.IslandExchange -file server.params


Then you launch your islands:

java ec.Evolve -file crete.params

java ec.Evolve -file minoa.params


If you're running crete and minoa on separate machines from
the server, you'll need to modify the parameter to tell them
where the server is:

java ec.Evolve -file crete.params \
    -p exch.server-addr=address.of.server.com

java ec.Evolve -file minoa.params \
    -p exch.server-addr=address.of.server.com