Implementation and Performance Analysis of the φ-Failure Detector

Failure detection is a fundamental building block for ensuring fault tolerance in distributed systems. However, providing accurate and flexible failure detection in off-the-shelf distributed systems is difficult. Practical solutions to failure detection rely on some adaptive mechanism to cope with the unpredictability of networking conditions. However, while they provide reasonably good accuracy, they also lack the necessary flexibility to provide failure detection as a system-wide service. In particular, traditional solutions take a “one size fits all” approach, which prevents them from simultaneously supporting several distributed applications with very diverse QoS requirements. In this paper, we present a novel approach to adaptive failure detection, called φ-failure detector, which addresses the flexibility issue mentioned above. We describe an implementation, and analyze its behavior over intercontinental communication links during several weeks. Our experimental results show that our failure detector compares well with other known adaptive failure detection mechanisms, with the considerable advantage that it provides virtually limitless flexibility.