Performance optimization method for sharing cache

Abstract

The invention discloses a performance optimization method for sharing cache, which adopts a technology of useless block elimination and low reusing block filtration to manage sharing cache resource, so as to enable the cache resource to be used fully and obtain higher performance. The technology adopts an algorithm based on counting to forecast the useless block and replaces the useless block as soon as possible, adopts a dynamic insertion and improvement strategy to filter the low reusing data, so as to keep the potential activity data as far as possible and prevent part of work set from being disturbed by data with low operating frequency. The experiment shows that the invention has higher performance relative to the traditional sharing cache management policy.

Description

technical field [0001] The invention relates to the field of computer hardware, in particular to a method for optimizing the performance of a shared high-speed cache. Background technique [0002] With the increasing speed gap between processors and memory, memory system design has become one of the key factors affecting computer system performance. Currently, multi-core processors generally provide fast access to recently accessed data through a large-capacity, highly associative last-level on-chip Cache (Last-Level Cache: LLC). Theoretically speaking, in order to achieve the highest Cache hit rate, LLC should adopt the Optimal replacement algorithm (Optimal replacement algorithm, OPT). The most recently used (Least Recently Used, LRU) replacement strategy or its approximate algorithm has been widely used in commercial processors. [0003] However, a large number of research results in recent years have shown that the performance gap between LRU and OPT in high associativ...

AI Technical Summary

Problems solved by technology

[0005] The traditional LRU replacement strategy usually puts the most recently hit data block on the MRU (Most Recently used) bit, and eliminates the data on the LRU bit. After some data blocks are replaced into the chip, they will no longer be referenced. Class data blocks are called useless blocks, which cause useless blocks to stay in the chip for a long time, and can only be replaced when they reach the LRU bit, which will greatly waste resources in the chip. Therefore, a good shared cache management The strategy should be able to predict and eliminate useless blocks; the traditional LRU replacement strategy manages the data blocks in the slice equally, but after some data blocks are replaced into the slice, they are only referenced a few times. It is a low-reused block, which will occupy the space of a high-reused block if it stays in the chip for a long time, resulting in a waste of on-chip resources. Therefore, it is necessary to be able to filter low-reused blocks while inserting and upgrading hit data blocks;

Images