That is very interesting issue you raise. I understand you are interested in fine-grained distributed memory applications: I have experimented recently with Proactive.  While learning curve might be steep (which is not helped by rather unresponsive mailing list for Proactive), after you work out initial problems with deployment it is great fun to work with. The only draw back is that after you transform your agents into Proactive ActiveObjects and create environment based on Proactive Domains, there is no Mason left (Proactive takes care of scheduling). It is possible to have large scale simulation deployed to cluster based on Proactive and small scale desktop Mason one that share 90% of code, in particular you won't have to change neither your agent code or environment code, just things that deal with instantiation and visualization.

Proactive comes bundled with tool called IC2D (Interactive Control and Debugging of Distribution), which is great help when you want to monitor or improve performance of your simulation.  I have heard good things about Network Attached Memory solutions like Terracotta, but have not yet tried them out: they look interesting though. Check out proceedings of last AGENT conference, there are some papers that talk about those issues..

What scale of simulation are we talking about that a decent 8 core desktop with plenty of RAM is not enough?



On Dec 6, 2007 7:59 PM, Glen E. P. Ropella < [log in to unmask]> wrote:
Hash: SHA1

What toolchains do y'all use for parallelizing your MASON simulations?
Note that I'm not talking about multi-processor machines but clusters or
loose collections of networked machines.

- --
glen e. p. ropella, 971-219-3846,
The only way we can win is to leave before the job is done. -- George W.

Version: GnuPG v1.4.6 (GNU/Linux)
Comment: Using GnuPG with Mozilla -