集群协作缓存机制研究

计算机集群中的节点使用内存一般不均衡,往往有些节点使用太多内存,而其他节点又有较多的空闲内存.为了改进集群操作系统,将集群节点的内存作全局分布的资源使用,我们首先提出一个内存互操作高速缓存方案:通过使用集群范围内的内存作文件高速缓存,从其他节点的高速缓存中读文件,可以避免很多低速的磁盘访问,改进集群文件系统的总体吞吐量.然后利用我们提出的缓存页面代替策略GCAR来支持这种内存互操作的高速缓存方案.该算法与CAR相比,对缓存中被"经常"使用的页面的管理粒度更细,更适合集群协作缓存的计算环境.实验结果表明,GCAR对本地缓存的命中率比CAR略好,在集群协作缓存下能取得更好的缓存命中率.

Study of Interactive Caching in Cluster Computing

In generally there will be some nodes in a cluster that are short of memory while others have idle memory wasted.How to improve the cluster system to support the use of cluster-wide memory as a global distributed resource? In this paper,at first we propose an interactive caching scheme,which allows the operating system to avoid many expensive disk accesses by using cluster-wide memory for file reading,so improves the overall throughput of cluster file systems.Further we also integrate cache replacement policy GCAR with this interactive caching scheme,GCAR has smaller granularity to manage "frequency" pages of cache than CAR,so it is more suitable for cluster computing environment.Experiment results show that GCAR can improve cache hit-ratios both at local cache and interactive cache in cluster computing.