Theory and Practice of Chunked Sequences
Sequence data structures, i.e., data structures that provide operations on an ordered set of items, are heavily used by many applications. For sequence data structures to be efficient in practice, it is important to amortize expensive data-structural operations by chunking a relatively small, constant number of items together, and representing them by using a simple but fast (at least in the small scale) sequence data structure, such as an array or a ring buffer. In this paper, we present chunking techniques, one direct and one based on bootstrapping, that can reduce the practical overheads of sophisticated sequence data structures, such as finger trees, making them competitive in practice with special-purpose data structures. We prove amortized bounds showing that our chunking techniques reduce runtime by amortizing expensive operations over a user-defined chunk-capacity parameter. We implement our techniques and show that they perform well in practice by conducting an empirical evaluation. Our evaluation features comparisons with other carefully engineered and optimized implementations.
Umut A. Acar, Arthur Charguéraud and Mike Rainey
ESA: European Symposium on Algorithms, September 2014