Koza says his recent 1 knode Beo cluster is well below saturation with a single FastEthernet NIC (and they probably have bottlenecks in their tree of switches), so an GA running on a sea of xDSL machines across country might be not that silly after all. (If any distributed.net people are listening, please email me).
A "GP-crashme" client could be created around http://www.bochs.com/ as point of departure. Once coredumping rarely, the system could run natively. Alas, the Windoze versions will always have to run in a sandbox, thus limiting search space speed.
John Holmes writes:
> I was under the impression there was a team working on such a networked
> parallel computing infrastructure. However, I thought they were centered at
> distributed.net and from what I just saw of that site, not too much has been
> accomplished in the past year. If there is not such a project going on, then
> it certainly seems there should be. Any volunteers to get one going?