针对当前OrangeFS中所存在的性能瓶颈,讨论分布式文件系统的优化方式,分析文件大小以及文件布局对分布式文件系统吞吐率的影响,提出一种采用共享内存与消息传递相结合的本地数据访问模型,减少了OrangeFS对网络带宽的使用,提高了文件的读写性能.实验结果证明本地数据访问模型能够提升152%的文件二次读性能、105%的文件写入性能.
For the bottleneck of performance in the OrangeFS,this paper discussed the ways to optimize the distributed file systems and analyzed how different file sizes and data layouts affected the system throughput.This paper proposed a hybrid data flow model which used both the shared memory and message passing to optimize the local data access,reducing the use of network of the OrangeFS,increasing both the file write and read bandwidths.The experimental results show that the local data access model is effective which can achieve up to 152% file re-read performance improvement and 105% file write performance improvement.