cientParallel Sorting Algorithm Richard

A parallel sorting algorithm is presented for storage-eecient internal sorting on MIMD machines. The algorithm rst sorts the elements within each node using a serial sorting algorithm, then uses a two-phase parallel merge. The algorithm is comparison-based and requires additional storage of order the square root of the number of elements in each node. Performance of the algorithm on two general-purpose MIMD machines, the Fujitsu AP1000 and the Thinking Machines CM5, is examined. The algorithm is suitable for implementation on special-purpose parallel machines, e.g. parallel database machines. There is a large literature on parallel sorting { see, for example, 1, 2, 9] and the references given there. Many of these papers have dealt with the problem from a theoretical point of view, neglecting issues which are important in a practical implementation of a parallel sorting algorithm 3, 7]. This paper describes a fast, practical parallel sorting algorithm which has been implemented on several MIMD machines and used in applications such as speech recognition and text retrieval. We aimed for an algorithm with the following properties. 1. Speed. The algorithm is competitive with the fastest known algorithms. 2. Good memory utilisation. The number of elements that can be sorted is close to the number that can be stored in the memory of the machine. 3. Flexibility. No restrictions are placed on the number of records to sort or the number of processors. 4. Determinism. The algorithm does not use a random number generator. 5. Comparison-based. The only operation used on keys is binary comparison.