Sliding-window compression for PC software distribution

Abstract

Summary form only given, as follows. We study the use of the LZ77 sliding window algorithm to compress PC files for distribution. Since the files need to be compressed only once and expanded many times, one can afford to use a complex compression scheme but must maintain a simple and fast expansion phase. In the experiment we allow the copy-length to be as large as 210 K which is the window buffer size used; this allows the expansion program to run even in the old PC/XT and compatibles. A suffix tree is employed to search for the longest matched length so that the search time is independent of the window size. We employ two methods to encode the displacements and copy-lengths. The first uses a modified unary code to encode the quantities (LZU) while the second method uses Huffman codes to encode them (LZH). Results and comparisons with UNIX's COMPRESS and the PC archive program LHA are tabulated.