Active replication is one of the main approaches for ensuring fault-tolerance. Yet, most implementations of active replication either require special hardware, or modify the programming model. Clearly, the former incurs high costs, implies being behind the technology curve, and may require special operating systems. On the other hand, altering the programming model means that applications must be modified in order to enjoy fault-tolerance, which imposes high software development costs and special programming skills. With the proliferation of virtual machine based languages, it becomes appealing to consider implementing fault-tolerance at the virtual machine level. This provides transparency to the application, yet can run on any commodity hardware and OS. In this work we explore the possibility of providing fault-tolerance by replication at the virtual machine of Java. Specifically, the main problem to implement fault-tolerance by replication is ensuring consistency of the replicated state. We first analyze the sources of non-determinism in a virtual machine, and propose methods of eliminating them. We then describe our implementation of these ideas within an open source JVM called JikesRVM, and present detailed performance measurements. Finally, we list the limitations of our approach and open challenges.
Roy Friedman is a senior lecturer with the Department of Computer Science at the Technion - Israel Institute of Technology, now on sabbatical at INRIA Rennes in France. He holds a B.Sc. and D.Sc. from the Department of Computer Science at the Technion. Dr. Friedman's research interests include dependable distributed systems, group communication, QoS of middleware, and mobile computing. Dr. Friedman was also a researcher in the Computer Science department at Cornell University, and is one of the Founders of PolyServe Inc.
Sprecher: Dr. Roy Friedman
Department of Computer Science at the Technion - Israel Institute of Technology
Wann: Dienstag, 03. Juni 2003, 15.00 Uhr (s.t.)
Wo: z-109 Universität Klagenfurt