parallel distribution of data

[Matt W]

Not so sure about my previous comments now.

Nichols, for the time being I think you'll have to run replicated:  if
they fit into memory replicated, the size of systems you're talking
maybe wouldn't gain too much in performance from the (expected
benefits of) distributed grids anyway.  I'll try and figure out what's
going on here, that allocation request is huge, but it requires
working out a way of simulating these systems without running the full
simulation.

Cheers,

Matt