Monitoring framework for stream-processing networks

In this paper we present a monitoring framework that exploits special characteristics of stream-processing networks in order to reason the performance. The novelty of the framework is to trace the non-deterministic execution which is reflected in i) the dynamic mapping and scheduling of network components at the operating system level and ii) the dynamic message routing across the network at runtime. We evaluate the efficiency with an implementation for the coordination language S-Net, showing negligible overhead in most cases.