On the Impact of Modelling Choices for Distributed Information Spread

We consider a distributed shuf¿ing algorithm for sharing data in a distributed network. Nodes executing the algorithm periodically contact each other and exchange data. The behavior of the algorithm is probabilistic in nature; a node chooses a random peer and sends a random subset of its local data. Moreover, the algorithm exhibits nondeterministic behavior; the order in which nodes initiate an exchange is not speci¿ed. For the shuf¿ing algorithm we build several formal models using the probabilistic model checker PRISM. Despite of the well known state-space explosion problem, we were able to model a network of up to 15 nodes. In addition, we implement two equational models in MATLAB, a discrete model and its continuous alternative, as well as the algorithm itself in the peer-to-peer network simulator PeerSim. By comparing different modelling frameworks, we further explore the impact of modelling choices, such as different scheduling policies and the notion of rounds. The evaluation of distributed protocols, especially gossiping protocols, is dif¿cult and a comparison of different evaluation techniques is greatly desired, since the evaluation techniques vary a lot and are based on different assumptions. The comparison of different models allowed us to discover hidden assumptions, which helps with the interpretation of the obtained results.