CURRENT BEOSIM PLAN:

Lazy Algorithm Modules

Class 1: Modules that know BSBW of job        

	Module 1: ()
Use the BW equation to identify the possible intervals of nodes that can
be co-allocated on a given link. ATTEMPT to co-allocated job into these
intervals, starting with the cluster that will accomodate the largest
partition of the job, else RETURN NULL.
	PRO: This module WILL NOT result in link saturation.
	CON: may not find solution, even when one exists.            
	
	Module 2: (Load-balancing)
Use the BW equation to identify the possible intervals of nodes that
can be co-allocated on a given link. Iteratively, ATTEMPT to co-allocate
the job one node at a time, as evenly as possible across the mini-grid,
else, RETURN NULL.


Class 2: Modules that don't know BSBW of job
	
	Module 1: (Largest available nodes first)
Discount all clusters with saturated links, of the remaining
clusters, co-allocate job starting with the cluster with the largest
number of available nodes.
	CON: This module may oversaturate links.

	Module 2: (Least saturated link first)
Discount all clusters with saturated links, of the remaining
clusters, co-allocate job starting with the cluster with the least
saturated link.
	CON: This module may oversaturate links.

	Module 3: (Chunking Big-Small)
Discount all clusters with saturated links, of the remaining
clusters, ATTEMPT to co-allocate job by placing the a "large"
chunck on a single cluster, and then place the remaining nodes 
on other clusters, else RETURN NULL.

	Module 4: (Load-balancing)
Discount all clusters with saturated links, of the remaining
clusters co-allocate the job, as evenly as possible, across the
remaining clusters.

The purpose of having these modules, is that if one of them fails
another module can be called to re-attempt allocation.

Implement the above algorithms for use in the CLUSTER 2004 
paper submission.

Finish IPDPS presentation (make new figures for some concepts)

COMMUNICATION MODEL VERIFICATION:

Create MPI program to test link saturation effects:
	for x iterations of outter loop
		for y iterations of COMPUTATION LOOP
			DO DUMMY COMPUTATION
		end COMP LOOP
		for z iterations of COMMUNICATION LOOP
			DO ALL-TO-ALL COMMUNICATIONS
		END COMM LOOP
	END outter loop

Get this running the nodes that are available to us 
on Adenine and C switch.