Experimental and Efficient Algorithms

Abstract

Algorithms for sorting large datasets can be made more efficient withcareful use of memory hierarchies and reduction in the number ofcostly memory accesses. In earlier work, we introduced burstsort,a new string sorting algorithm that on large sets of strings is almosttwice as fast as previous algorithms, primarily because it is morecache-efficient. The approach in burstsort is to dynamically builda small trie that is used to rapidly allocate each string to a bucket.In this paper, we introduce new variants of our algorithm: SR-burstsort,DR-burstsort, and DRL-burstsort. These algorithms use a random sampleof the strings to construct an approximation to the trie prior tosorting. Our experimental results with sets of over 30 million stringsshow that the new variants reduce cache misses further than did theoriginal burstsort, by up to 37%, while simultaneously reducing instructioncounts by up to 24%. In pathological cases, even further savingscan be obtained.