Semantically aware contention management for distributed applications

Abstract

Distributed applications that allow replicated state to deviate in favour of increasing throughput still need to ensure such state achieves consistency at some point. This is achieved via compensating conflicting updates or undoing some updates to resolve conflicts. When causal relationships exist across updates that must be maintained then conflicts may result in previous updates also needing to be undone or compensated for. Therefore, an ability to manage contention across the distributed domain to preemptively lower conflicts as a result of causal infringements without hindering the throughput achieved from weaker consistency is desirable. In this paper we present such a system. We exploit the causality inherent in the application domain to improve overall system performance. We demonstrate the effectiveness of our approach with simulated benchmarked performance results. © 2013 IFIP International Federation for Information Processing.