The Research On Memory Compression For Embedded System Based On Pre-decompression Mechanism

Memory is scarce resource, especially for memory constrained embedded system.More physical memory means more package cost, more power consumption, biggerPCB size. Besides, this issue becomes more serious under the memory requirementtrend that the average amount of memory required by applications has been grown50%~100%every year. One solution to this problem is memory compression whichcould increase the available memory of the system by compressing inactive memorypages. However, frequently performing compression and decompression coulddegrade the performance of the applications since the decompression occur in thecritical path of execution.Based on those observations, this paper presents a novel memory compressiontechnique for embedded system. The main contributions of this paper are listed below:Firstly, performance model for memory compression of embedded system isestablished. Based on this performance model, analysis of gain and cost is given.Secondly, a new memory compression technique for embedded system calledCMPD (compressed memory with pre-decompression) is presented. In order to avoidthe decompression latency, we design an adaptive pre-decompression algorithm whichis based on the principle of temporal locality. After that, the details of design ofCMPD is given, which include the selection of compression algorithm, adaptivecompressed memory area and memory allocator.Finally, the CMPD is implemented as a loadable module against Linux kerneland the implementation details are given. The experimental results indicate that,comparing to previous memory compression technique, CMPD could increase theavailable memory of embedded system with negligible penalty.