Tunable Replica Consistency for Primary-Backup Replication in Distributed Soft Real-time and Embedded Systems

In systems that use primary-backup replication for fault-tolerance, maintaining system availability after failures refers not just to ensuring the liveness of application functionality at a backup replica but also to ensuring that the state of the promoted backup matches that of the failed primary. Traditionally these availability criteria are realized in middleware through sophisticated algorithms that provide a certain level of replica state consistency, such as strong or weak. For distributed real-time and embedded (DRE) systems, the constant fluctuations in resource availabilities and application workloads, however, preclude a tight coupling to any single criteria for state consistency thereby forcing the need for new mechanisms that can tune the replica consistency algorithms at runtime in accordance with the operating conditions, and DRE system timeliness and availability requirements. This paper describes preliminary work in this space and highlights important challenges that must be addressed. keywords: real-time and availability, replica consistency, tunability.