Research On Hybrid Storage Model Based On Resistive Random Access Memory

In the era of big data,the scale of data is increasing day by day,and the storage system is faced with many challenges such as speed,capacity,power consumption,cost,scalability,non-volatile and endurance.The traditional dynamic random access memory(DRAM)is limited by its storage density and process size,and is incapable of dealing with massive information storage requirements.Compared with DRAM,the new nonvolatile memory(NVM)has the advantages of high storage density,low power consumption,and large capacity.It provides a direction to solve the problem of storage caused by large data growth.Resistive Random Access Memory(RRAM)has outstanding performance in all aspects,and it has certain representativeness.This research built hybrid memory based on RRAM.The content is summarized into three aspects.1?Build a hybrid memory simulator.In order to realize the characteristics of analog DRAM and RRAM memory devices,the structure of Gem5 and NVMain simulation systems and the relationship between various functional modules are studied.The memory model,the internal methods and functions used are focused on,and a hybrid memory simulation platform is constructed to provide a guarantee for follow-up research.2?Built a DRAM-RRAM hybrid memory model.A hybrid storage model is proposed,which introduces RRAM as the same level storage device based on the traditional memory structure.The RRAM can improve the capacity and reliability of the memory system for its high storage density,strong extensibility,non-volatile and other characteristics.We can distribute requests reasonably,and make use of the fast writing efficiency of DRAM to make up for the shortcomings of new memory write delay by constructing the hybrid memory controller.The performance of hybrid memory system is evaluated under benchmark test set PARSEC.The experimental results show that the performance of the hybrid memory system with DRAM and resistance memory as main memory is close to the system with DRAM as main memory and better than the system with RRAM as the main memory,which fully verified the effectiveness of the hybrid storage model proposed in this paper.Analysis of the factors that affect the performance of hybrid memory systems.It is found that the application's read-write ratio,hybrid memory capacity ratio will affect the performance of hybrid memory systems.3?The hybrid-memory page management strategy based on page write popularity.Considering the defects of RRAM writing latency and endurance,a hybrid memory strategy page management is proposed.According to the request access to the memory page,the write page is defined in hot and cold state,the hot write page is preferentially stored in the DRAM,a reasonable migration mechanism is constructed,and the number of writes on the RRAM is reduced.Experiments show that the cool and hot division of the writing page can effectively improve the system performance,and significantly reduced the system latency for write operation intensive applications.