Real-time Prediction Model of Cache Miss Rate Based on Local Memory Access Characteristics

Resource management algorithms that aim to reduce competition in shared caches need to obtain real-time process allocation on the cache. However, as the complexity of the processor architecture increases, the accuracy of cache allocation information becomes increasingly difficult to guarantee. This paper analyzes the causes of changes in the cache allocation of a process, introduces the concept of the cache miss rate change intensity, and proposes a real-time prediction model for the cache miss rate based on the local memory access characteristics of the process. The model can predict the cache allocation and cache miss rate of the next stage in real-time based on the number of executed instructions of the process and the number of caches misses. The prediction model proposed in this paper can quickly adjust to ensure prediction accuracy and real-time performance in the case of sudden changes in the cache miss rate. Through the test, the average accuracy of the cache miss rate can reach 95.7%.