机群文件系统中一种文件迁移的客户端决策

File migration client side decisions in clustered file systems

为降低机群文件系统中文件访问的平均响应时间,提出了一种能反映机群节点上文件访问频繁程度的热量模型,在此模型的基础上给出了一种文件迁移的客户端决策,用于决定在非重负载情况下,单个文件在机群中的理想存放位置。模拟对比实验的结果显示,采用该策略,在总负载恒定的情况下,在4、8节点机群上文件访问的平均响应时间分别降低了21.6%和26.3%,在平均负载恒定的情况下,在4、8节点机群上文件访问的平均响应时间分别降低了22.2%和16.8%,结果表明该策略可以有效地降低文件访问的平均响应时间,提高系统的可扩展性,从而提高机群文件系统的性能。

A model was developed to describe the file access rate on clustered nodes. The model was used to analyze file migration decisions in the client to decide the ideal location for an individual file for light loads. Simulations showed that the strategy can reduce mean response times by 21.6% with 4 nodes and by 26.3% with 8 nodes with equal total loads. The mean response times were reduced by 22.2% with 4 nodes and by 16.8% with 8 nodes with equal average loads. Therefore, this file migration strategy not only reduces the mean file access response time, but also has good scalability, so that it can improve the performance of large clustered file systems.