MonoScope: Automating network faults diagnosis based on active measurements

Network faults, such as router failures, cable outages, and configuration errors, could seriously affect network performance. In this paper, we use network faults very loosely, including those that will yield unfavorable end-to-end network performance, such as packet reordering and suboptimal routes. Diagnosing network faults on end-to-end paths is a very challenging problem, because it generally involves other domains. Even if it can be done, the process is very time consuming, because multiple sources of data, which are scattered in different places, are needed for such diagnosis. In this paper, we consider the problem of making the fault diagnosis as automatic as possible. Based on coordinated active measurement from a set of end systems, we propose a procedure of detecting network faults and identifying their locations. Although this procedure cannot be fully automated for the time being, we show that some of the components could be automated, and we are automating them in a preliminary system called MonoScope. We demonstrate the efficacy of this procedure through several real-world cases that we have encountered in our four years of network monitoring experience.