Research And Implementation Of Replication Scheme In Distributed System

With the rapid development of computer network, the performance and availability of distributed application system is facing a great challenge. The load of system may be very heavy because of large numbers of users and vast concurrent visit, then the reliability and performance of system can not be ensured in distributed application system. Replication could improve the performance of system, provide high availability and fault tolerance, and it is the key technology to ensure the validity of distributed application system.The China Central Radio and TV University Academic Affairs Management System is a large distributed application system. In order to improve the performance of this system, an asynchronous replication scheme based on MOM is designed and implemented. The consistency of replicas is an important common demand. Because of the complexity of network, how to guarantee the consistency is a key issue to implement replication scheme. According to the limit of consistency, it can be divided into several types. Eventual consistency is weak, but it has well performance.There should be a reliable and asynchronous communication between the replicas in order to guarantee eventual consistency. The Message Oriented Middleware (MOM) not only supports asynchronous communication, but also can improve the reliability of communication and the performance of the end points. In order to unify the interfaces of diversified MOM, Java Message Service (JMS) defines a standard of asynchronous communication based on Java. Therefore, it can take advantages of MOM based on JMS when used in asynchronous replication scheme.Adopting the MOM of JBossMQ, JMS and Message Driven Bean (MDB) in this scheme, a loosely coupled asynchronous communication is implemented. The PTP and Pub/Sub two modes are both adopted in delivering message and it guarantees the eventual consistency of replicas. Updating conflicts, Lamport logical clock, sequence of updating message and failure of replica are analysed and solved in the scheme. After putting this replication scheme in practice,using LoadRunner to test the performance of this system and analyzing the result of testing, it is proved that the scheme can improve the performance of the system.