Memory Hierarchy Based Performance Review of Sorting Algorithms

Sorting is known to be one of the fundamental computing problems which continue to attract a great deal of research. It is evident that instruction count requirements of the key sorting algorithms are well understood. However, the cost of executing a sorting algorithm is influenced by factors apart from instruction count such as cache miss and page faults as well. There exist studies of sorting algorithms from memory hierarchy perspective, but these studies are stale and losing relevance in the context of modern computer architecture. We have experimented with Quicksort, Merge sort, and Heapsort to assess their respective performances in memory hierarchy of modern computing environments. Inferences derived from the outcome of the experiments have been presented in this paper. Empirical evidence confirms that Quicksort is still an excellent algorithm from memory hierarchy point of view. Also, it has been inferred that Merge sort is poor on large records because its page fault count is too high. Heapsort was found competitive enough to be considered as an algorithm of choice on large records in the view of its lower page fault count. I. Introduction Sorting is known to be one of the fundamental computing problems which continue to attract a great deal of research [1-6]. Though instruction count requirements of the key sorting algorithms are well understood, however, in virtual memory systems, the cost of running an algorithm depends on other performance factors as well. Literature suggests that sorting algorithms execute slow if these sorting algorithms cost too many cache miss or page faults. The total running time is comprised of the time needed to handle paging traffic, cache miss penalty and actual processor time. It is noticed that when the number of records to be sorted, N, is small, paging traffic is low, and the execution time of the sorting algorithm is essentially governed by processor time. Also, if number of records is small, only a couple of pages will be used and hence choice of sorting algorithm will not have any impact on page fault count. As N grows larger, the paging traffic increases, and when a very large number of records are sorted, the paging time can be far more important than the processor time. Also, if N is large, the memory requirements of different sorting algorithms may vary considerably. Thus, the efficiency of a sorting algorithm depends both on the inherent time complexity of the algorithm …