Performance analysis and optimized operation of structured overlay networks

The ongoing process of globalization leads to a huge demand for highly scalable applications. The Peer-to-Peer (P2P) technology enables an arbitrary large number of users to participate and contribute to distributed services like content distribution or collaboration. With P2P, the intelligence is shifted from centralized instances to the edges of the Internet. This shift is supported by the widespread availability of powerful terminals and broadband networks. In P2P systems, these end terminals create a powerful overlay network, which is highly scalable since new users automatically add new resources to the system. The main goal of P2P overlays is to e ciently nd and share distributed resources among users in the Internet. However, the functionality of a deployed system heavily depends on the maintenance of its overlay topology. A disruption of this overlay structure can cause anything from degraded performance or limited functionality up to the point of a total collapse of the system. Thus, current research tackles these problems on di erent level, like optimized self-organization schemes, e cient overlay maintenance and data transfer, and short lookup delays. Structured P2P networks arrange resources in the system according to a well-de ned overlay structure. Thereby, proactive routing ensures that each node knows about a certain part of the current overlay. As a result, querying arbitrary resources can be performed within a limited number of hops. The most common approach to realize the necessary overlay structure are distributed hash tables (DHTs). In this thesis, we concentrate on the scalability and robustness of these overlay structures, as well as the lookup of resources stored in the overlay. Based on a detailed simulative and mathematical performance analysis of structured P2P protocols, we show that DHTs scale well with the number of participants, just as they are designed to address the scalability problem inherent to unstructured P2P networks. However, P2P systems operate in dynamic environments and peers frequently join and leave the overlay (churn). Therefore, maintaining a correct overlay structure in rapidly changing scenarios is di cult. In this context, it is important to adapt the con guration of design parameters to the actual network size and churn rate. Based on these results, we evaluate various existing modi cations and extensions to