Some potential ideas for Will's PhD proposal.

1.  What we have done thus far
2.  k-ary n-cube for describing local application comm. patters
3.  additional network topologies

4.  varying job mixtures, larger number of clusters with varying sizes
	-- run some simulations where coallocation is required (many jobs 
	larger than can be run on a given cluster
5.  PDF for having random job BSBW's that converge to a mean that is 
	controlled by a parameter so graphing is still possible

6. Add phases to jobs that do comp -> comm -> loop, for more dynamic nature

7. Address the potential to have Coven expose communication information to the
scheduler before and during runtime that could be used in making scheduling
decisions. Perhaps a static data-flow analysis of the program to tell how much
communication will be taking place, and amoung which nodes of a job.

8. Use Coven's checkpointing capability to justify the ability for the
scheduler to use runtime job migration as a means by which to move a comm.
bound co-allocated job to a single cluster for example.

9. Address issues of scale, larger clusters, more clusters, Earth Simulator.

10.  Address node heterogeneity. This is important for IEEE TPDS SI.

Tenative Schedule:

Proposal
	Early November (Before SC'ing)
Submit to IEEE TPDS
	Early Jan
Defense
	Early May