TODO

Renaming...

Shouldn't initMsgEnv() be called as part of the constructor of
AlpsKnowledgeBrokerMPI? 

Should there be default encode() and decode() methods? Or should we force the
user to pack everything? Or should we just do it with bufferes like BCP?

Why is there a bestBound in ALPS? Why is the default -1.0 (assuming
positivity?)

Why is the function called receiveSizeNode()?

Shouldn't exploreSubTree be called from within hubMain() or workerMain()? No,
probably not... 

In exploreSubTree, why not use the constructor that takes a subtree as an
argument. That will eliminate the setSubTree() call and the
setupKnowledgePools() call. 

AlpsAllParam should probably be renamed.

workerMain: largeSize, mediumSize, smallSize?

How does MPI_Bcast work? It seems to be called identically for both sender and
receiver. 

Can we find a way not to check for a solution every iteration in workerMain()?

Need to separate out worker and hub??

Should knowledge broker really be driver? I suppose yes...

AlpsKnowledgeBase no longer used?

============
12/03/05

* Remove data pool. Move parameter and model data to model.
* Move solution pool from broker to model.
* Add function to compare the goodnesss of two solution.
* Remove largeSize, etc.