SyncProbe: Providing Assurance of Message Latency through Predictive Monitoring of Internet Paths

SyncProbe improves the end-to-end predictability of distributed systems by providing applications with a real-time estimate of the maximum expected message delay (upper bound on communication latency) for network paths. The upper bound is adjusted over time in response to the monitored network latency and serves as a real-time assurance of synchrony. We deployed SyncProbe on PlanetLab and assessed its performance with respect to violations, duration of synchrony, upper bound cost and recoverability. Experiments revealed that SyncProbe is successful in providing upper bound estimate in real-time for a variety of paths. The estimated upper bound can be utilized for more formal assessment of information related to timeouts, ordering of events or knowledge about global states. We describe the design and methodology of SyncProbe and discuss various issues related to its performance.