Supervised peer-to-peer systems

In this paper we present a general methodology for designing supervised peer-to-peer systems. A supervised peer-to-peer system is a system in which the overlay network is formed by a supervisor but in which all other activities can be performed on a peer-to-peer basis without involving the supervisor. It can therefore be seen as being between server-based systems and pure peer-to-peer systems. The supervisor only has to store a constant amount of information about the system at any time and only needs to send a small constant number of messages to integrate or remove a peer in a constant amount of time. Thus, with a minimum amount of involvement from the supervisor, peer-to-peer systems can be maintained, for example, that can handle large distributed computing tasks as well as tasks such as file sharing and Web crawling. Furthermore, our concept extends easily to multiple supervisors so that peers can join and leave the network massively in parallel. We also show how to extend the basic system to provide robustness guarantees under the presence of random faults and also adaptive adversarial join/leave attacks. Hence, with our approach, supervised peer-to-peer systems can share the benefits of server-based and pure peer-to-peer systems without inheriting their disadvantages.