Performance comparison of distributive and mergesort as external sorting algorithms

Abstract

The performance of two basic external sorting algorithms, distributive sorting and mergesort, is compared in an environment where even the main memory usage involves a cost. Performance is measured by total execution time and main memory space-time integral. For optimal behavior, both algorithms prefer a small block size and a similar order of external sort. Their memory requirement is similar during the external phase, but during the internal phase distributive sorting requires a larger working space. For small records, the optimal behavior of distributive sorting is obtained with less external passes and its space-time integral is smaller. For large records, the number of passes at the optimal point of mergesort is similar or even less than that of distributive sorting, resulting in a smaller space-time integral. In all cases, the performance of distributive sorting degrades more mildly around the local optima.