
28 Apr
2008
28 Apr
'08
8:02 p.m.
Right. However, what we would like to accomplish here is to move away from speculation over how poorly one method performs over the other, and gather some good empirical data that covers all (or at least the most prevalent) usage patterns.
Fortunately, my proposal sorta covers this type empirical investigation. Part of it is to look at other some other graph libraries and see how they provide various aspects of the interface. My guess is I'll end up producing a bunch of little generic kernels that can be composed to provide variations on searches (e.g., traversal, find first, find all, etc.). Andrew Sutton asutton@cs.kent.edu