Non-Volatile Memory Based Cache Replacement Policy

Currently, most computer systems consist of Dynamic RAM (DRAM) as main memory and Hard Disk Drives (HDDs) as storage devices. Due to DRAM’s volatile nature, it suffers from data loss in the event of power failure or system crash. With rapid development of new types of non-volatile memory, it becomes promising that one of these technologies will replace DRAM as main memory in the near future. Considering non-volatile memory as main memory in our future systems, some mechanisms designed for DRAM based system should be modified to fully explore the advantage of the non-volatile characteristic of main memory. In today’s DRAM based systems, newly updated pages will have to be flushed to HDD in every 30 seconds. In the future non-volatile memory based systems, any updated pages can be kept longer in main memory without the urgency to be flushed to HDD. Therefore, this opens opportunities to investigate new cache replacement policy. In this paper, we investigate and propose a new non-volatile memory based cache replacement policy that giving priority to re-group and synchronize long consecutive dirty pages to take advantage of HDD’s fast sequential access speed and nonvolatile property of main memory. Specifically, with the help of a fixed or dynamic Threshold, we synchronize and evict qualified consecutive dirty pages out of cache, instead of evicting singular page at the least recently used position. We evaluate our scheme with various I/O traces. The experimental results show that our proposed cache replacement policies shorten system I/O completion time more than 10x and improve cache hit ratio varied from 0.5% to 10% when compared with DRAM based LRU policy.