RDMA Based Design And Implementation Of Distributed Memory File System

With the advent of the Internet era,a large amount of data has been generated from various social activities.Therefore,the storage of data and the efficient retrieval and retrieval of the required data at the same time has become a problem to be solved.Distributed file system is an effective solution.The emergence of Non-Volatile Memory(NVM)provides new options.NVM is a high-performance storage device;its biggest feature is that it does not lose data when power is off,and its performance is similar to memory;it has the advantages of byte addressability and read and write latency at the nanosecond level.Researchers have made many attempts for NVM;basically,it is done on a single node.However,it is difficult for a single node file system to expand the storage capacity.The storage capacity is very limited and cannot meet the needs of society.For NVM,how to make full use of its high performance and can be applied to distributed file systems is a subject worthy of research.With the continuous improvement of the performance of the CPU and I / O system,the transmission performance of Ethernet can no longer meet the requirements.The traditional TCP / IP network protocol stack is very complicated,which makes the network transmission speed and memory access speed have a large difference,which results in the overall performance of distributed file systems that rely on Ethernet for data transmission.RDMA(Remote Direct Memory Access)technology stands for Remote Direct Memory Access,which is generated to solve the delay of data processing between the client and server during network transmission.Therefore,combining NVN and RDMA to improve the performance of distributed file systems is an important task.This article will introduce an RDMA-based non-volatile memory storage file system(RDMA NVM File System,RNFS for short).RNFS has the ability to quickly expand the storage capacity of NVM,and uses the high performance of NVM to achieve fast file data through RDMA Read and write.This paper mainly conducts the following research: 1)Architecture design that supports node expansion.The RNFS file system uses a master-slave structure.The master node is responsible for metadata and management of the slave nodes,and the slave node is responsible for providing file data storage space.2)RDMA-based file data read and write process.Use the RoCE protocolto achieve efficient RDMA file data read and write operations.3)Storage space allocation mechanism based on bitmap linked list.The bitmap linked list is used to manage the allocation and usage of NVM in RNFS,which combines the advantages of bitmap space saving and fast allocation of linked lists.In order to ensure the read and write performance of large data blocks of files,the mechanism of pre-allocating continuous physical space is used to accelerate the allocation process and the multi-threaded background collection is used to speed up the data block release process.4)System disaster recovery function.For a complex distributed file system,node failure and master host performance bottlenecks are very common problems.RNFS implements data backup and dual-system hot standby functions to improve system stability.Finally,a test experiment was performed on the RNFS file system.