Research On The Performance Optimization Method Of Hardware And Software Based On The System Simulation

The expanding application scale poses a huge challenge to the underlying hardware of the computer.However,the performance gap between the CPU and the main memory is becoming more larger with the gradual development of the underlying hardware of the computer,and the problem of "scissors difference" is gradually becoming obvious.Cache has become the best choice to alleviate the problem of"scissors difference".However,Cache has been developing for decades,and its performance has not been a major performance breakthrough.The traditional structure of Cache is gradually unable to adapt to new application requirements.On the other hand,the larger scale of applications make developers rarely consider the storage characteristics of the underlying Cache when they develop the program,resulting in inefficient program execution.The traditional structure of Cache and the increasing diversity of program data access have formed a pair of sharp contradictions,and the specialization gap between hardware and application has been increasing.In view of the above contradictions,this thesis comprehensively investigates the structure and working principle of the traditional Cache,and deeply analyzes its structure and working principle,and points out the shortcomings.On this basis,for the specialization gap between on-chip Cache hardware and application,the method is proposed in software and hardware respectively.For the diversity of data,this thesis proposes a separation mechanism based on reading and writing attributes.The implementation method is to optimize the storage of data according to the Cache characteristics before the data is loaded into the cache through two stages of marking and separation;The shortcomings of the traditional Cache structure can be summarized as the consistency processing of all the data of the program in the Cache storage.For this problem,this thesis proposes a separation mechanism based on the frequency attribute.By counting the access frequency of the Cache block,this thesis conducted the research on separation and recombination mechanisms.Through the attribute-based hardware and software separation mechanism,the cache hit rate of the Cache can be effectively improved,thereby optimizing the performance of the Cache.In addition,this thesis proposes a neighbor cache block writing back strategy for the problem of frequent replacement of Cache blocks.In order to accurately evaluate the impact of the proposed method on Cache performance,this thesis exploited GEM5 simulator to simulate the basic computer architecture and deployed Linux system experiments to verify the effectiveness of the proposed method.The experimental results show that the improvement effect of RHR based on the method of reading and writing attribute separation is stable at about 5%,the improvement effect of WHR is stable at about 9%,the maximum can reach 10.2%,the improvement effect of OHR is stable at about 6.8%,and the maximum value is up to 7.39%;.When the main memory coverage is 10%,The separation mechanism based on frequency attribute has the best improvement for the cache hit rate which is up to 19.58%,and the lower the cache coverage,the more obvious the improvement effect.The adjacent cache block writing back strategy reduces the memory writing back rate of the Cache by an average of two percentage points.Through the Linux system experiment,the effectiveness of the method is indirectly reflected.Therefore,the hardware and software Cache performance optimization method proposed in this thesis can effectively improve the performance of Cache.