Using Memory Compression To Improve System Performance

For decades, CPU speeds have continued to double every 18 months to 2 years, but disk latencies have improved only very slowly. Disk latencies are five to six orders of magnitude greater than main memory access latencies. Programs that run entirely in RAM benefit from improvements in CPU speeds, but the runtime of program that page is likely to be dominated by disk seeks, and may run many times more slowly than CPU-bound programs.Due to the great latencies of I/O incurring by swapping, we can buy more RAM to improve the system performance, but by this way, the system costs increase. In order not to increase the costs of current system, we take advantage of memory compressionThe idea of memory compression is to hide the disk latencies by storing swapped out page frames in a compressed form, but still in physical memory. On a subsequent page fault, the page can be quickly decompressed and supplied to the application program.The actual implementation of the compressed memory system is in the form of a loadable device driver module. This module provides support for compressed in-memory caching of block devices. Compared with the previous implementation, our system can be dynamically loaded and un-loaded.The test works approved that the design function is satisfied and the various applications gain various improvements from the driver.