Prediction-Guided Performance Improvement on Compressed Memory Swap

Abstract

Due to ever increasing demands for memory size, compressed memory swap technique has been widely deployed in many consumer electronics. Although reducing data size effectively extends available memory, it inevitably brings computational overhead. Also, the effectiveness of this technique highly depends on the compression ratio. If there is a significant amount of incompressible data, the compression only brings unnecessary overhead without any benefits. In this paper, we address this problem by skipping the compression of incompressible pages in an efficient manner. We propose a novel compression predictor which quickly and accurately estimates whether a page is compressible or not. The experimental results show that our predictor can improve launch time by 29.5% on average with 97.4% accuracy.