graph centrality benchmark

[StefanKarpinski]

See this discussion of Julia vs. Python vs. C++ performance on a graph centrality computation. We should turn this into a perf2 benchmark and make sure we get within 2x of C++. Sample code:

• C++: https://gist.github.com/3362265
• Python: https://gist.github.com/3353764
• Julia: https://gist.github.com/3353761

Sample data: http://www.udacity.com/view#Course/cs215/CourseRev/1/Unit/273003/Nugget/292002.

[JeffBezanson]

The single biggest problem seems to be iterating over Sets. The way it wraps Dict iteration is probably part of it. Dict could also be better in general; ObjectIdDict is still much faster.

I also notice that strip is kind of slow, although it doesn't contribute significantly to this particular program. split is important as always.

[JeffBezanson]

Just noticed this is now significantly faster than it used to be.

[westonplatter]

@JeffBezanson is this issue finished? Anything remaining?

[ViralBShah]

In my quick benchmarking with the codes from the gist, it appears that we are a bit faster than the C++ code, and much faster than the python code.