The Research And Implementation Of Optimization Of PMFS

With the rapid development of the internet,data is booming and various applications put forward higher requirements on the storage system's capacity,delay,bandwidth and so on.Emerging non-volatile,byte-addressable memories(NVMs)like phase-change memory and 3D-XPoint can make data persistent at main memory level instead of storage.This new architecture offers opportunity to create the next-generation system of storage.Since the read/write latency of NVMs is very low,the overhead of software in a NVM system may determine the whole performance.PMFS is a file system specifically designed for NVM,reducing the overhead by bypassing the page cache,block generic block level and device driver.In this work,we find some bottlenecks of PMFS and then do some optimization.The contributions of this work are summarized as below.First,by evaluating the performance of PMFS under various workloads,we show how much PMFS performs differently with traditional disk file system.The results show that PMFS's performance in read and write is much better than PMFS,however,under workloads of practical application,the advantage is not significant.Second,we find PMFS has an undesirable characteristic: each directory operation(create,open or delete)requires a linear search of the entire directory files.This overhead becomes the bottleneck of whole system in some circumstance of particular workloads.To solve the problem,we add a directory index for PMFS.The experiments show the optimization accelerates creating files by 12 x and improves the system bandwidth by 27.4%.Finally,we find PMFS degrades seriously when appending files as the number of threads increases.To solve the problem,we optimize the allocator and the experiments show the optimization improve the bandwidth by 145% when 8 threads do concurrent appending.