Analysing cache effects in distribution sorting

Abstract

We study cache effects in distribution sorting algorithms. We note that the performance of a recently-published distribution sorting algorithm, Flashsort1 which sorts n uniformly-distributed floating-point values in O(n) expected time, does not scale well with the input size n due to poor cache utilisation. We present a two-pass variant of this algorithm which outperforms the one-pass variant and comparison-based algorithms for moderate to large values of n. We present a cache analysis of these algorithms which predicts the cache miss rate of these algorithms quite well. We have also shown that the integer sorting algorithm MSB radix sort can be used very effectively on floating point data. The algorithm is very fast due to fast integer operations and relatively good cache utilisation.