Research On Recovery Management In Distributed Transaction Monitor

Transaction technology is a key technology to ensure the consistence and reliability in information systems. The most frequently used flat transactions are atomic operation sequences with ACID (atomicity, consistency, isolation and durability) characteristics.Recovery management is the essential problem of transaction process. Recovery manager is very important for maintaining the ACID property of the transaction. And distributed transactional recovery aims to ensure the durability of the objects on different servers and the atomicity of the failures. For distributed transactions, the challenge of recovery management is to recover the failures occurring during the Two-Phase Commit Protocol.Transactional recovery plays an important role in database applications and in distributed network environment. By far, the research on recovery management in database systems is relative thorough. The ARIES algorithm by C. Mohan from IBM Research Center is the classical algorithm applied in the databases systems. However, the research on the recovery management in distributed transactions is still very significant and the distributed computing environment in J2EE brings higher requirement for transaction recovery.In this thesis, we studied several key issues of distributed recovery management.First , we develop the design and implementation of a JTS Recovery Manger-JRecovery. JRecovery provides recovery service for ISTX1.0 which is a distributed transaction monitor. JRecovery is able to efficiently deal with the failures of distributed systems.We also proposed several optimizations in the log management and information searching. JRecovery implements server kinds of logs and realizes the virtual log. The virtual log is scalable and easy for other implementations of logs. Moreover, the virtual log reduces the burden of the log reader and writer but it also brings some side effects. During recovery management, the transaction information search becomes troublesome since the details of log files is masked. JRecovery tries to optimize the search procedure. We give the algorithms for the restoration of active transactions' list and the transaction status searching and the optimizations of the two algorithms.