Study On Parallelization Of MBR Codes Based On FUSE

In order to solving the problem of high storage overload,erasure code is introduced in the storage system.Erasure code can significantly reduce the storage overload of the system,but the network bandwidth of the system will be severely congested when we use it to repair some failed node.Therefore,it is an urgent problem how to read and write data quickly,and repair failed node quickly while reducing the network traffic of the system.To solve above problems,this thesis designs and implements a P-MBR system supporting distributed storage based on FUSE,and realizes the parallelization of MBR codes.Specifically,it includes the following aspects:(1)we study the parallel reading performance of MBR codes,mainly include normal read and degraded read;by using slice technology in every block,we rewrite the read function of FUSE,make it possible to read data block from two node by slice when reading file every time,thus to improve the system performance of read.(2)We study the parallel write performance of MBR codes.Taking advantage of the feature that each block of MBR codes is distributed on two different nodes,we rewrite the write function of FUSE,and make it possible to write data block to both nodes at the same time,so as to improve the parallelization of the writing process of the system.(3)In order to improve the system's repair performance,we speed up the process of reading data and writing data by cache layer and MBR codes layout;in addition,multithreading technology is also used to speed up the parallelization of system's repairing,thus improving repair process.Experimental result show that the MBR codes has a small improvement in reading and writing performance compared with RAID 5 and RAID 6;a great improvement in degraded reading and repair performance,the degraded read performance is improved by about 70% and the repair performance is improved by about 40% and 60% respectively,a fast repair speed and a low network traffic.