SFT-ERF: Scalability and Fault-tolerance in an Event/Rule Framework for Distributed Systems

 

In Distributed Systems (DSs) with Asynchronous, non-deterministic Reactive Components (ANDRCs), the timing assumptions are not valid and the output could be different, even if the same sequences of stimuli are input with the same initial state. In such systems, the availability and reliability are critical, so they must be fault tolerant and it can be achieved by replicating the ANDRCs. This research uses active replication with a middle-tier component that multicasts incoming-events to all replicas, and detects and suppresses duplicated outgoing-events that were posted by each replica; it also uses a semi-active replication approach where only one (“the leader”) is able to post events, while the others listen to the leader to keep consistency, having each replica detecting/suppressing duplicated outgoing-events. The Event-Rule Framework (ERF), which is a framework for developing DSs with ANDRCs, was used as a test bed for this research. The performance analysis shows linear execution-time curves in all test scenarios. Therefore, proposed solutions were proven to be feasible, and their performance results were proven to be acceptable.