Monitoring pairwise interactions to discover stable wormholes in highly unstable networks

Abstract

Users of large-scale testbeds often need a group of nodes with a reasonable level of stability to execute applications and experiments. Although monitoring the stability of nodes themselves is certainly part of the solution, it is important to classify and select groups of nodes according to their ability to communicate among themselves. In this work we call such groups of nodes "stable wormholes", and describe strategies to find those wormholes based on monitoring end-to-end pairwise interactions. Data acquired is used to find five different types of wormholes, each with a different stability pattern. The system was implemented in PlanetLab. Extensive experimental results are reported evaluating the proposed strategies. A comparison with another tool that selects nodes based on node stability alone is also presented. The execution of a MapReduce application shows that nodes selected with the proposed strategy ran the application significantly faster. © 2012 ICST Institute for Computer Science, Social Informatics and Telecommunications Engineering.