Choose a suitable problem, perform a brief literature search, select three recent and competing state-of-the-art algorithms, and implement them on shared memory and/or distributed memory platform. Study their performances, perform fine tuning, and write an implementation based paper describing the problem, major approaches, selected algorithms with justifications, implementations details and performance tuning, plots, discussion of plots, and possible future work and your conclusions. A plus would be to come up with your own algorithm (either theoretical or practical).
A good implementation paper would be of publishable quality.