Performance of network and service monitoring frameworks

The efficiency and the performance of management systems is becoming a hot research topic within the networks and services management community. This concern is due to the new challenges of large scale managed systems, where the management plane is integrated within the functional plane and where management activities have to carry accurate and up-to-date information. We defined a set of primary and secondary metrics to measure the performance of a management approach. Secondary metrics are derived from the primary ones and quantifies mainly the efficiency, the scalability and the impact of management activities. To validate our proposals, we have designed and developed a benchmarking platform dedicated to the measurement of the performance of a JMX manager-agent based management system. The second part of our work deals with the collection of measurement data sets from our JMX benchmarking platform. We mainly studied the effect of both load and the number of agents on the scalability, the impact of management activities on the user perceived performance of a managed server and the delays of JMX operations when carrying variables values. Our findings show that most of these delays follow a Weibull statistical distribution. We used this statistical model to study the behavior of a monitoring algorithm proposed in the literature, under heavy tail delays distribution. In this case, the view of the managed system on the manager side becomes noisy and out of date.