Monitoring and characterizing application service availability

Abstract

Reliable detection of global application service availability remains an open problem on the Internet. Some availability issues are diagnosable by an administrator monitoring the service locally, but far more may be identified by monitoring user requests (ie. DNS / SSL misconfiguration). In this work we present Proddle, a distributed application layer measurement framework. The application periodically submits HTTP(S) requests from geographically diverse vantages to gather service availability information at the application layer. Using these measurements we reliably catalog application service unavailability events and identify their cause. Finally, analysis is performed to identify telling event characteristics including event frequency, duration, and visibility.