Design And Implementation Of Checkpoint And Garbage Collection Of Log-Structured File System

With the development of Internet and big data,the gap between size of data and ability of processor become bigger and bigger.Traditional computers often use hard disk as backup device,whose speed of write and read is slow.The architecture which is based on computing is hard to meet the requirement of performance and time of big data,therefore in memory computing which is based on memory becomes a solution to big data.The non-volatile memory can be used as normal memory of computer and has many features,such as fast speed of reading and writing,byte-addressable,lower energy consumption.The research on non-volatile memory mainly focus on programming model,simulation system and storage system.Based on Linux kernel,Non-volatile memory file system implements all interface provided by Virtual File System of Linux and can be used in various applications without modification.However,most of them focus on performance of reading and writing instead of data consistency and optimization of the hybrid use of DRAM and non-volatile memory.Like the traditional file system,they use fsck to restore after system crash,which is too slow to satisfy the requirement of big data.We propose a new file system Hybrid Memory File System(HMFS)that guarantee the consistency of storage system with low IO cost.In this paper,we present the design and implementation of checkpoint of HMFS and the algorithm of garbage collection of memory file system.In the last,we do experiments to show performance of file system in checkpoint and normal operations.