Research Of Page Management Algorithm For Hybrid Memory Architechture

Recent years have visited the rapid development of VLSI, which makes the processor’s performance significantly improved. In addition, the parallel, distributed, cloud computing computer give more power for information processing. However, with the emergence of big data, various problems that exist in storage system has become increasingly prominent, such as DRAM which serve as main memory has high power consumption, low integration. This issues have become obstacles for the development of modern computer systems.Non-volatile memory (NVM), for example, PCM, FeRAM, ReRAM, STT-RAM, etc. Due to its’low static power consumption, high-density, non-volatile, high read speed, etc. has provided a new opportunity to upgrade modern computer memory systems. However, compared to the traditional DRAM memory, NVM also has shortcomings that can not be ignored, such as the limited write count, high dynamic power, and long write delay. For Giving full play of the advantages of various storage media, this paper uses PCM and DRAM to cambine hyrbid main memory, and use NAND flash as storage memory, configured computer hybrid storage system. Specifically, this paper researches the page cache management in hybrid memory.First, ignore the external storage, this paper researches the page cache management in PCM+DRAM hybrid memory. To make full use of the PCM’s superior static power consumption and high integration while the DRAM’s short write latency and low write power consumption, this paper provides an adaptive page management policy called AIMR (Adaptive page Insertion, Migration, and Replacement). Based on the effective recognition of hot page, write hot page from cold ones in a workload, AIMR consists three algorithms, including page insertion, page migration, and page replacement.Second, considering the NAND flash storage, this paper researches the page cache replacement algorithm. Compared to the traditional magnetic disk, there is no seek time inside NAND flash. Therefore, NAND flash has high random read speed. However, it’s random write delay is longer than magnetic disk because of the garbage collection inside it. To improve the I/O performance of NAND flash, this paper proposes an adaptive page replacement policy called APTS (Adaptive page Replacement algorithm by combing Temporal and Spatial locality in NAND flash based system).Third, In order to fully and accurately verify the page cache management algorithm for hybrid storage system, this paper integrate AIMR and APTS together. The method is AIMR works for the PBR in APTS. This paper uses full system architecture simulator GEM5 which is widely used in academic research to test the algorithm after the fusion of AIMR+APTS. We attached NVMain+DRAMSim+McPAT+FlashSim to GEM5. Experiments show that our proposed page management algorithm AIMR and APTS can greatly improve the performance of the whole hybrid storage system and extend its lifetime.