Providing Reliability in Replicated Middleware Applications

Abstract

Problem statement: Data inconsistency is raised in actively replicated environment due to non-determinism in the applications that defeats the purpose of replication as a fault-tolerance strategy. Approach: We proposed an efficient framework RTC which ensured determinism among the replicas in fault tolerance middleware applications. This method exploits the technique of statically analyzing the application source code of client and identifies the variables and system calls which lead to non-deterministic state in the replicas. The source code consists of non-deterministic variables and system calls which are identified and set the flag field. The client request consist of flag field and the service request, which is sent to all the servers through time stamp based replication protocol (TSP) that facilitate the multiple clients and the request is sent to the servers. The distributed coordination method was initiated if necessary; otherwise send any one response of the servers to the client by duplicate removal. Distributed coordination which involves, the selection of a primary replica based on the time stamp value. It is responsible for taking all non-deterministic decisions. The state of the primary replica was updated to all other replica connected asynchronously to maintain consistency. Results: We evaluated our technique by increasing the contamination percentage of non-determinism and increasing number of replicas. Conclusion: The method suggested by us reduces the communication and synchronization overhead which was proved through implementation. We evaluate our technique for the active replication of servers using micro benchmarks that contain various sources of non-determinism. Multi-threading, system call, shared I/O and random ( ).