EVALUATION OF COMPRESSED CACHING

Abstract

CPU speeds have been increasing by about 60% every year and disk latency is improving by less than 10% a year. Virtual memory systems use disks as slow, very large storage for program data when main memory is full. Compressed caching is the division of main memory into two pools: uncompressed cache of pages and compressed cache of pages in compressed form. By compressing some of main memory, the effective memory size is made larger and disk access can be avoided. When uncompressed cache is full, pages are evicted into compressed cache. Similarly when compressed cache is full, pages are evicted to disk. Compressed caching has been evaluated through trace-driven simulation which is a log of events, taken from real program execution, to allow a simulator to mimic the behavior of a system. Trace-driven simulator can use a reference trace to track the movement of pages from one memory level to another, recording the cost that would be incurred to fetch and evict those pages. Simulation of compressed caching is carried out on three different systems namely Pentium Pro, Pentium III, Pentium IV. Compressed caching provides better results on fast machines like Pentium IV with compression and decompression times of about 0.023 ms and read or write speeds of about 1.5 ms, under high and normal memory pressures. The results for compressed caching are not satisfactory even for Pentium IV when large pool of free memory is available than required by the program. In case of slower machines like Pentium pro with read and write speeds of about 5.3 ms and compression and decompression times of about 0.6ms, compressed caching is not suitable under high, normal and low memory pressures. The results also depend on the foot print of the programs or number of memory pages required by the programs. Programs with large foot prints are benefited by the compressed caching