[Spread-users] what configuration for a LAN of 150 machines dynamically groupedby 2 or 3 ?

[John Lane Schultz]

Philippe Reinquin wrote:
> - Suppose I have a network of 150 machines (I modified the 
> MAX_PROCS_RING parameter).

Spread out of the box will almost surely have serious problems running with such 
a large ring.  In order to support that many daemons you would have to (a) 
increase the timeout latencies by quite a bit (which means larger latencies to 
detect network+daemon failures) and/or (b) change Spread's base ring protocol to 
account for such a large ring (warning: for *advanced* Spread programmers only). 
  As an example, you could change Spread to multicast its tokens (more network 
overhead)  as well and to reset the ring timeouts whenever a daemon sees the 
token make meaningful progress from one daemon to another.

Alternatively, you can reduce the number of daemons in the system down to a size 
that we recommend of between 30-40 daemons and have clients on the 150 machines 
connect remotely to an appropriate daemon.  I'd recommend you to go this route.

If all of your daemons are in the same LAN (i.e. - can IP multicast/broadcast to 
one another) then putting all the machines in the same Spread_Segment is the 
best way to go.

Cheers!

-- 
John Schultz
Spread Concepts LLC
Phn: 443 838 2200
Fax: 301 560 8875