Distributed Breve simulations
Has anyone developed distributed Breve simulations? This would seem to enable more realistic simulation, since the processing can be distributed.
I`d be interested in hearing about your experiences with this capability.
Thanks, Alan
working on it...
This is a hot topic for breve development right now, and something I'm actively working on -- see the thread here, for example: http://www.spiderland.org/node/3378
Generally speaking, there is no out-of-the-box way to automatically distribute simulations across multiple machines, but there is support for networking that has allowed users to implement their own distribution to various degrees. This can range from sending actual objects from one simulation to another, to sending and receiving parameter data via a dedicated server.
One important exception to the lack of out-of-the-box networking is Genetic Algorithms written with the Push programming language -- these simulations have support for automatic networking built-in, and it can provide a big performance benefit.
- jon
breveCreatures "net breeding" code working ?
Around this subject, Jon, Can you please tell us if that migration feature is working or not in the screensaver ?
Looks like there is some possibility meant for it in the xml file : migrationHosts, migrationPercent, etc, but I can't get it to work (don't know which format to use to fill in the migrationHosts list).
Would be a starting base to develop on regarding distributed breving :)
Thanks for your attention
Distributed GP migration is
Distributed GP migration is fully working and tested for the PushGP class (which implements its own GP, separate from GeneticAlgorithm).
It appears that it should be working with GeneticAlgorithm as well, but I haven't used it as extensively.
To try it out, start up the same GeneticAlgorithm based simulation on two computers, and set the migrationHosts for each one to be a list containing the hostname of the other:
On computer1:
ga set-migration-hosts to { "computer2" }.
On computer2:
ga set-migration-hosts to { "computer1" }.
- jon
I had thought that there was
I had thought that there was a Beowulf cluster that could divide up Breve processing.
I think they were doing
I think they were doing multiple runs across multiple machines to collect huge amounts of data -- not doing a single simulation distributed across many machines.
- jon