The Measurement And Optimization Of Parallel File System In Mass Storage System

Mass storage system (MSS) is designed for store mass data. This system can scale horizontally, that is, MSS can expand the both performance and capacity of the system without increasing the complexity of system management. Parallel file system is comprised of a group of nodes, and it finishes the computing task more quickly by means of communication and cooperation of nodes. The parallel file system plays a great role in the high performance mass storage system, so it has important significance in high performance and cloud storage, and has been wide concerned and extensive researched.Testing on parallel file system is divided into several categories: conformance, functional, stress, performance and speculative, of which the performance test is the focus. The performance of parallel file system is assessed by throughput and input/output operations per second (IOPS), the former is to measure the speed of big file reading and writing, while the latter is the measurement of small file reading and writing speed.Common optimization performance strategies used in parallel file system involve data prefetching, cache, write-behind, striping and so on.Cache in file system refers to the use of buffer cache to save the contents of recently accessed files, to improve the efficiency of file access. Data prefetching is to read data into memory before actual use of the data. Write-behind aggregate small files into a big file to reduce the number of disk operations. Striping is the technique of segment logically sequential data, such as a file, in a way that accesses of sequential segments are made to different physical storage device.In this thesis, we study the test methods in parallel file system, and the strategies to improve the performance. The study includes: parallel file system test standards and methods, and study to improve file system performance test methods, and take GlusterFS's optimizing policy for example. The result shows, the cache scheme in GlusterFS improves 10MB/s read speed, while the prefetch scheme in GlusterFS enhance 15MB/s read speed.