The project, timetable and notes and such
	It's going...ok
	Most people/teams have a chance of getting it to work

Complexity of solutions:
	Complex solutions are often better
	We'll try to avoid stereotypes by using concrete examples
	Suspension, tanks, and airplanes
	Sometimes a simple algorithm is best
	Why is a complex algorithm not just as reliable?
		Each piece has a probability of a bug (debug it?  An OS works...)
		Bugs are not all a problem in a specific piece of code
		I played a duet with my daughter yesterday
		Interactions between n components...
	Sometimes a simple solution will NOT work!
		History of operating systems...
		There's a reason for AI!
	Don't overestimate problems

Map/Reduce paradigm
	Google used to use it, and had a proprietary system with that name
	Programming paradigm, not really a single technology

Large shared memory (Cray XMT and such):
	Convenient!
	Traditional "supercomputer" setup
	Tends to have limited size
	EXPENSIVE (in all caps)

Clusters:
	Not expensive in all caps
	Can be very large

Cluster with special network:
	Better than a regular cluster
	Expensive (capital E) depending on how "special"

Mesage Passing (MPI):
	Latency, network size vs. bandwidth

MapReduce and Hadoop:
	map operation (available in Python!)
	reduce operation (I won't say anything isn't available in Python)
	data locality
	wikipedia has some good "standard" examples
	Data structure in the center:
		Hash table?  Tree?  List?  Often some sorting required
		Sort of depends what you're using it for

Checkpointing with Map/Reduce:
	Straightforward 
	
What model does a GPU fit into?
	They have shared memory
	But, many independent units that cannot easily confer with one another
	

Applying this to GPU processing of images:
	Can we figure out the average contrast variance of an image?
	Of course...