Conversion of NSP DAGs to SP DAGs

Random multithreaded NSP task graphs cannot be modelled in SP-based parallel programming models. In this report we present a new algorithm for converting NSP DAGs to SP DAGs. Our analysis establishes an upper bound of a factor of 2, on the increase in critical path length(W∞). However, the total work (W1) remains the same. We also evaluate an implementation of our algorithm and perform subsequent analysis with real-world models of task graphs and random task graphs which provide encouraging results as compared to previous research. Specifically, we prove the correctness and analyze computational complexity of our algorithm followed by a heuristic based variant for further performance enhancements.