BOUNDED SIZE DICTIONARY COMPRESSION: RELAXING THE LRU DELETION HEURISTIC

Abstract

The unbounded version of the Lempel-Ziv dynamic dictionary compression method is P-complete. Therefore, it is unlikely to implement it with sublinear work space unless a deletion heuristic is applied to bound the dictionary. The well-known LRU (least recently used) strategy provides the best compression performance among the existent deletion heuristics. We show experimental results on the compression effectiveness of a relaxed version (RLRUp) of the LRU heuristic. RLRUp partitions the dictionary in p equivalence classes, so that all the elements in each class are considered to have the same "age" for the LRU strategy. Such heuristic turns out to be as good as LRU when p is greater or equal to 2 (RLRU2). RLRU2 is slightly easier to implement than LRU in addition to be more space efficient.