A preemptive view change for fault tolerant agreement using single message propagation

Abstract

The paper presents a proactive approach for failure detection. In our previous work [5], we assumed a trustworthy Transaction Manager, TM, amenable to the job of view creation, detection of faulty primary as well as backup replicas and to evacuate them from the transaction processing system. In the end, TM initiates a view in case it detects faulty primary or faulty replica. The TM provides an efficient failure-resiliency in the protocol; however, it also introduces the possibility of single-point failure. To eliminate the reliance on single TM, we propose a protocol that distributes the responsibilities of a transaction manager among 3f+1(f are faulty) replicas and results in a distributed Transaction Manager (DTM). The article attempts to limit the failure detection time to an optimum value, i.e., single message propagation time between any two nodes. © 2011 Springer-Verlag.