Research On Write Optimization Of Non-voliate Memory File System Based On DRAM Cache

The emerging non-volatile memory(NVM)has both low latency read/write and byte addressable memory features,but also has the disk's non-volatile features.In recent years,the rapid development of NVM technology makes research based on the NVM file system becoming a central issue.To avoid the double-copy overheads caused by copying data among the user buffer,the OS page cache and the storage layer,NVM-aware file systems directly copy data between the user buffer and the NVM storage.However,one of the major drawbacks of the NVM storage device(like PCM)is that the write speed is low.Direct access for all file operations can result in suboptimal system performance.Regarding the issue above,this paper uses DRAM as the data cache of NVM by using the asynchronous write back strategy to copy DRAM cache data into NVM to improve system write performance.In the design of a multithreaded scheduling policy,this paper combines the multicore characteristics of processors to dynamically adjust asynchronous writeback threads based on the number of user threads that perform write requests and the number of CPU cores to take full advantage of CPU resources to become a high-performance file system HBFS based on NVM.In addition,in order to reduce the software overhead of sharing resources such as global space lock between multiple threads,HBFS divides the NVM and the DRAM cache space into multiple resource pools for management to be allocated in parallel by different threads.HBFS also provides consistency-consistent guarantees.The transaction mechanism ensures the integrity and atomicity of file operations with fine-grained logs used for metadata and copy-on-write strategy for data.After an experimental test,comparing to current NVM file systems,HBFS has improved the I/O throughput rate by 67%,write speed by 70%,and verified the consistency of HBFS through SQlite.