Throughput Improvement Of Distributed File Systems

With the rapid developments of computing technology,the problem of I/O bottleneck becomes more and more severe. It is aggravated because of the demand of high performance file accessing ability in the distributed environment. Improving the performance of distributed file systems is crucial to the whole systems.Throughput is an important metric of the systems'performance. This paper works on the issue of throughput improvement of cluster file systems.First,this paper introduces two classical file systems—NFS and PVFS.Then,based on comparing and analyzing the throughput of these two systems,this paper points out the key factor which influences the throughput mostly.Finally,the author presents his idea of multi-component lookup which is used to improve the throughput of cluster file systems. This method proves very useful according to the testing results before and after this implementation.In Chapter 1,this paper introduces the general characteristics,several design issues and caching technique of cluster file systems.Then the background for this paper is discussed.Chapter 2 introduces the structure,control flow,protocol and implementation of NFS.Chapter 3 focuses on the structure,control flow and implementation of PVFS. In chapter 4,after presenting his method of classification of time spent in distributed file systems,the author turns to the testing and the analysis of the throughput of NFS and PVFS.In this chapter,the key factor which influences the throughput mostly is discussed.Also in this chapter,the author points out the shortcomings of the design and implementation of PVFS.According to the conclusions of Chapter 4,the author describes a design and implementation of multi-component lookup,which is used to improve the throughput of distributed file systems. The performance before and after this implementation has also been compared.This paper ends with the conclusions and future research directions which are given by the author.