CS 224 W Project Final Report CUDA Implementation of Large Graph Algorithms Group

Running SCC graph algorithms on large datasets can be a time-consuming task, and we spent the quarter investigating methods of parallelizing this task using CUDA. For very large graphs, too much time can be wasted by not parallelizing the graph algorithms, and we want some of the insights from our experiments to be used to speed up common graph analysis tasks. We initially started by implementing breadth-first search on the GPU, but then quickly moved on to implement a parallelized SCC algorithm. We then benchmark our parallelized SCC graph algorithm against a non-parallelized version of Tarjan’s algorithm to find strongly connected components.