Study On The Key Technology Of Log-Based Recovery In Main-Memory Data Management

Main-memory data management technology,especially main-memory database captures wide attention.Depending on its high concurrence and throughput,within low latency,main-memory database is various kinds of situations.Along with dramatic increases in performance and decreases in cost of computer hardware,hot fields and prosperous future are filled in main-memory database research.Log-based recovery,which is one of the most important research orientations,is a set of problems accompanied by volatile memory.The problem,that non-volatile storage result in stagnate in memory and CPU for its slow I/O speed,is now the issue that needs to be addressed urgently.Meanwhile,there is no special platform for log-based recovery research.Main contributions are as follows.1.For the special platform issue,we design and implement a simulation platform called RecoS.RecoS aims at an implementation of recovery sub-system of main-memory database.It uses cluster substrate to simulate realer data storage,and developed interfaces for variety of recovery strategy.We also adapt ARIES and command logging on the platform,which represents physical and logical logging respectively,focusing on their recovery process and some important details.Failure recovery and checkpointing is also described as relevant course,and some unique feature is discussed caused by main-memory.2.For the problem of slower I/O,we tried to make some progress start with minimissing log size and optimize log commit.We proposed a set of strategies based on encoding and group commit,making use of redundant information in this log group.Dictionary and indirect encoding is used to compress.A new algorithm is presented with a threshold based on indirect encoding.Relevant experiments is done on RecoS.It proves that encoding method shorts sizes of log group effectively and it saves memory spacement in other words.Furthermore,group commit reduces I/O time obviously.