Tolerância a falhas em java através de comunicação em grupo.

Abstract

The intense use of Distributed Systems in different kinds of applications demands more reliability of these applications. The systems should be prepared to tolerate faults, i.e., a system should not interrupt its operation in the presence of some components (hardware or software) and data faults. Active replication is usually used when one aims at building such high available and fault-tolerant services. Some Group Communication Systems already offer support for the development of fault tolerant distributed applications. However, most of those systems are not portable, a very important property in distributed systems. In this context, the Java language has become widely used in Distributed Systems in the last years, specially due to its portability and facilities for the development of distributed applications. Nonetheless, Java provides no support for the development of fault-tolerant distributed applications which can continue to function properly despite component failures. This paper describes an approach for fault-tolerance in Java which can meet the requirements of active replication. In order to achieve that, an extension to the iBus package designed by Silvano Maffeis [MAF96] has been developed and implemented. The developed system, named iBusTF (fault-tolerant iBus), added new group communication properties required by active replication : total order delivery and atomic membership. The approach adopted has the advantage of only using Java resources, keeping total compatibility with the iBus system.