已有的LRFU(Least Recency Frequency Used)自适应算法在实际应用中根据经验调整λ值,缺乏对访问局部性强弱的量化分析,因而其可适用的访问模式有限.该文首先建立基于K阶马尔可夫链(K→∞)的局部性定量分析模型,在访问过程中根据统计信息实时量化局部性特征.然后以此分析模型为基础设计自适应替换算法LA-LRFU(Locality-Aware LRFU),随着访问特征的变化动态调整参数λ.最后应用Trace仿真对算法进行测试.实验结果显示,针对多种访问模式,LA-LRFU均可显著提高Cache命中率;在由多种访问模式构成的具体访问过程中,LA-LRFU能比现有的各类LRFU自适应算法更合理地调整参数λ.
Network storage provides high scalability, availability and flexibility for storage systems, and is widely applied to many fields. Particularly, I/O performance is of great significance. Its application is wide and expanding rapidly. I/O performance has already become the bottleneck of the whole performance of computer systems for a long time, and under the condition of the present computer technology, I/O performance optimization method looks especially important. In the paper, I/O performance model was analyzed based on the combination of quasi birth, death process and queuing model, and then solved the model. A number of important related performance indicators and the relationship between them were given. By the way of example, this method can show the I/O performance more accurately. Finally, we got some useful conclusions, which may be used to evaluate network storage performance, and are the basis of confirming I/O scheduling strategy.
