对象存储中基于高斯分布的分层缓存淘汰算法

Layered Cache Elimination Algorithm Based on Gausssion Distribution in Object Storage

在分布式存储系统中,引入分层缓存技术是优化系统读写的重要方法。目前分层缓存技术多使用LRU及其改进算法管理缓存空间,该方式虽然一定程度上改善了缓存性能,但有限的缓存命中率也使其成为性能瓶颈。针对此问题,本文研究聚焦于分布式对象存储在云计算场景下的应用特点,设计了一种基于高斯分布的淘汰算法。算法基于高斯分布的形态特征设计淘汰规则,规避了LRU及其改进算法采用频率估计概率的误差影响。通过仿真实验结果表明,在用户访问符合高斯分布时,本文提出的基于高斯分布的分层缓存淘汰算法能有效提高缓存命中率。

Hierarchical cache is an important method for accelerating data reading and writing in distributed object storage.At present,hierarchical caching uses LRU and its improved algorithm to manage storage space.While this approach improves performance to some extent,its limited cache hit ratio also make it a performance bottleneck.In order to solve the problem of limted performance,this paper analyzes the application characteristics of distributed object storage in cloud computing scenarios,and based on the analysis the new elimination algorithm is designed.The algorithm designs the elimination rules based on the morphological features of the Gaussian distribution,and avoids the error impact of the LRU and its improved algorithm using the frequency estimation probability.The simulation results show that when the user accesses the Gaussian distribution,the proposed algorithm can effectively improve the cache hit ratio.