Evaluation of a Performance Prediction Tool for Peer-to-Peer Distributed Computing Applications

In the search of new architecture for distributed computing, peer-to-peer is studied as a new way forward. But, its intrinsic properties like the absence of centralized topology or the dynamic reorganization of the network make it difficult to reach high performances. Furthermore, it is not trivial to execute applications on an existing testbed which brings together a sufficient number of nodes. That is why it is necessary to use a simulation environment which will be used to bypass bottlenecks and to correct the parts of the applications which slow down the execution time. In this context, we have proposed P2PPerf: a simulation tool which aims at predicting performance and the execution time of a distributed application before its finalization. This article presents some experiments conducted with P2PPerf to evaluate its accuracy. The emphasis is put on the use of real execution platforms with the Grid'5000 platform and with the Wrekavoc tool.