Re: agents for distributed computation

[bwelch]

The term 'agents' is frequently a fuzzy term. Generally, I think of agents as independent workers that communicate with either each other or a 'mother' process. Other sources define them differently.

I see PBS on CPAN, a Portable Batch System. It's a queuing system and lets one spawn off many 'agents' to work on tasks independently. I've used systems similar to PBS to manage agent systems for bioinformatics applications. In those, thousands of agents are created, work on a piece of data, then either communicate that they've finished and die, or request more work. One application created 17,000 agents on a cluster and completed in around six hours. Note that optimization of granularity is important. Before optimization, the 86,000 agents were created and it took six days to complete.