Exploiting Content Locality to Improve the Performance and Reliability of Phase Change Memory

摘要

With the explosive growth in data volume, the I/O bottleneck has become an increasingly daunting challenge for big data analytics. The outstanding energy efficiency and scalability characteristics of Phase Change Memory (PCM) make it a potential, attractive alternative to DRAM and traditional storage devices. However, PCM's slow write performance and weak write endurance are the two major weaknesses that prevent its wider applications. Moreover, the slow processing workflow of write requests also causes significant contention for and interferes with read requests, thus affecting the system performance. In this paper, we propose Content Aware PCM (short for CA-PCM) that employs a lightweight data deduplication module to exploit the content locality in memory accesses. CA-PCM is able to effectively reduce write traffic to PCM by removing unnecessary duplicate writes and also substantially extend the lifespan of the PCM device. Our trace-driven simulation results show that CA-PCM improves the performance and reliability significantly.