shrink / expand for ad hoc compute nodes

[AndySomogyi]

Hi,

I'm exploring using Converse / Charm++ for a dynamic compute environment where nodes can be added and removed at run-time.

I know at the least, I'd have to write a new Char++ load balancer, but not really sure what kinds of changes Converse would need. Presumably something in "src/arch/netlrts" has a table that maps PEs/ nodes / sockets, but haven't found this yet, and not really how amenable this structure would be to runtime changes.

I've looked at the existing shrink expand in: "examples/charm++/shrink_expand", and not sure if it works.

Anyways, let me know if you think Converse and possibly charm++ could be a good basis for a dynamic compute environment.

Thanks