Load-aware Adaptive Failover for Middleware Systems with Passive Replication

Supporting uninterrupted services for performance-sensitive distributed applications operating in resource-constrained environments is hard. It is even harder when the operating environment is dynamic and processor or process failures and system workload changes are common. Fault-tolerant middleware for these applications must assure high service availability and satisfactory response times for clients. Although passive replication is a promising fault tolerance strategy for resourceconstrained systems, conventional passive replication solutions are nonadaptive and load-agnostic, which can cause post-recovery system overloads and significantly increase response times. This paper presents Faulttolerant Load-aware and Adaptive middlewaRe (FLARe), which enhances conventional passive replication schemes in three ways. First, its client failover strategy is load-aware, i.e., failover targets are selected at runtime based on current CPU utilizations to maintain satisfactory response times and alleviate CPU overload after failure recovery, and adaptive, i.e., failover targets are proactively adjusted in response to failures, system load fluctuations, and resource availability. Second, its client redirection strategy handles resource overloads that stem from multiple failures and workload fluctuations. Third, FLARe enables effective dissemination of failover decisions and manages CPU utilizations transparently to clients. Empirical evaluations on a distributed testbed demonstrate how FLARe efficiently uses available system resources and maintains satisfactory response times for clients when recovering from failures.