WAYS TO THINK ABOUT PARALLELIZING AN ALGORITHM:

(your mileage may vary)

- what is the data dependency structure of the problem?
- what is each processing unit (PU) doing?
- what data are in each PU?
- how many time clicks are required?
- how many PUs are involved at each time click?
- how can I keep all the PUs busy all the time?
- how can I usefully throw more hardware at the problem?