A Production Suite for Failure Detectors

Designing a failure detection algorithm, implementing it, and assessing its performance is extremely complex. The behavior of a failure detector (FD) implementation varies with the evolving conditions of the network, and its initial parameter settings also have a huge impact. In this paper, we introduce a novel production platform as a means of systematizing the design and performance assessment of FDs. It provides a language dedicated to the definition of FDs and uses pre-acquired real-world traces to emulate network conditions. Our platform simplifies and accelerates the design and testing of new FDs, and allows to compare their respective strengths in multiple environments.