Fast And Instantly Recoverable Storage Engine On NVM

Currently,emerging non-volatile memory(NVM)is gradually taking people's attention.Due to the combination of low latency,persistence,large capacity and byte-addressable,database system can run upon NVM-only storage architecture.In this environment,some novel logless database system came into being.Because of the elimination of logs,these systems can immediately recover after a sudden failure.However,under the current computer architecture,these systems need a large amount of synchronizations to ensure data consistency,which leads to a severe performance penalty.My contributions are as follow:1)Review the related work of NVM-based storage system,including the basic concept of NVM,NVM programming interface and synchronization,NVM-based database and so on.2)Implement a Fast and Instantly Recoverable Storage Engine(FIRSE),which runs upon NVM-only storage architecture.The first design philosophy of FIRSE is to divide the data into critical data which must be always consistent and rebuildable data that can be recovered,reduce the number of synchronizations by ensuring only the consistency of critical data in daily operations.On the other hand,the system can recover instantaneously through asynchronous recovery mechanism.3)Perform a complete evaluation of FIRSE and two other representative persistent data structures,including both independent performance evaluation and end-to-end performance evaluation by integrating them into the high-performance in-memory database Redis.The results show that FIRSE performs well under most workloads.Instantly recoverable is the most outstanding feature of FIRSE,which is 3-4 orders of magnitude shorter than other data structures in recovery time.Last but not least,FIRSE is also better than the other two data structures in NVM footprint and wearing.