6.895 Final Project: Serial and Parallel execution of Funnel Sort

The speed of a sorting algorithm is often measured based on the sheer number of calculations required to sort an n element array. However, in practice the sheer number of calcualtions is not the only factor in determining the running time of the sorting algorithm. For the purposes of the paper, we analyze the affect of cache/main memory memory transfers on the “Wall-clock running time” of a sorting algorithm. Analysis covers an implementation of the cache-oblivious, cache-efficient algorithm Funnel sort. Finally, this paper explores parallelizing the Funnel sort algorithm. We find it difficult to determine the number of memory transfers that Funnel sort uses in execution, but is still outperformed by Quick sort by approximately a factor of 4 when the array to be sorted fits entirely in main memory. In addition, we find that Funnel sort shows moderate promise as a parallel algorithm.