On Performance and Cache Effects in Substring Indexes

This report evaluates the performance of uncompressed and compressed substring indexes on build time, space usage and search performance. It is shown how the structures react to increasing data size, alphabet size and repetitiveness in the data. The main contribution is the strong relationship shown between time performance and locality in the data structures. As an example, it is shown that for a large alphabet, suffix tree construction can be speeded up by a factor 16, and query lookup by a factor 8, if dynamic arrays are used to store the lists of children for each node instead of linked lists, at the cost of using about 20% more space. And for enhanced suffix arrays, query lookup is up to twice as fast if the data structure is stored as an array of structs instead of a set of arrays, at no extra space cost.