Abstract
Large-scale testbeds provide realistic environments for the experimentation and evaluation of new protocols and distributed applications. In order to be successful, these experiments must be executed on sets of nodes that present a reasonable level of stability, and it is important to ensure their reproducibility. In this work we describe strategies to select sets of testbed nodes based on monitoring information. The system is modeled as a stability graph in which the vertices correspond to testbed nodes and there is an edge between two vertices if their communication is classified as stable. We investigate the performance of different topologies embedded in the stability graph to run experiments on PlanetLab. Results show that the k-core outperforms the other strategies in terms of their impact on the performance and reproducibility of the experiments. A k-core is a maximal subgraph of G in which all vertices have degree at least k. The average execution time of distributed applications executed on a k-core was up to 59% lower, and the variation on the results obtained was reduced by up to 29% when compared to other alternatives.
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have