Simulationsbasierte Analyse und Entwicklung von Peer-to-Peer-Systemen

Peer-to-Peer (P2P) systems are distributed systems composed of up to millions of functionally equivalent entities (peers), which form P2P overlay networks on top of physical networks to communicate. The functionality of a peer is implemented by a P2P application which de nes the behavior of the whole P2P system. The equivalence of peers is realized by providing client functionality as well as server functionality. Implementing a P2P system with speci ed behavior is a di cult task because the behavior depends on many factors, such as the used P2P search methods and the underlying physical network. Some factors cannot be taken into account completely because of their complexity or unknown or not understood parts. For instance, the prospective user behavior may only be estimated based on observed data. When engineering complex, dynamic software systems such as P2P systems, simulation is often used to analyze the properties of these systems based on simulation models in an early development phase. With simulations in natural sciences, the separation of reality and (simulation) model is clear: the reality exists in nature, while the model exists as software within some computer system. When simulating software systems, this separation is not so obvious: the simulated model is itself a software system. With P2P systems, for instance, a simpli ed P2P system is modeled and simulated for predicting properties of real P2P systems. The new software engineering contribution of this work is the Peer Software Engineering (PeerSE) method, which allows a controlled transition from simulation models to real-world software systems. The method starts with a comparative analysis of simulation models for P2P systems and proceeds iteratively toward the experimental implementation in a laboratory setting and nally a real-world P2P system deployed in a target environment. The method includes a simulation model for P2P systems and a tool supporting the execution of simulation and laboratory experiments. Simulation is an essential part of the PeerSE method used to identify and to compare models ful lling given requirements. When an appropriate model has been found, model components can be reused and further re ned to implement a laboratory P2P system. To allow for a controlled transition of model components to laboratory components, the results of simulation and laboratory experiments are directly compared using the same metrics. The applicability of the PeerSE method has been successfully evaluated by analyzing and realizing a P2P system for distributed software development.