Affordable Access

Publisher Website

A robust and scalable peer-to-peer gossiping protocol

Publication Date
DOI: 10.1007/978-3-540-25840-7_6
  • Communication
  • Design


ap2pc.dvi A Robust and Scalable Peer-to-Peer Gossiping Protocol⋆ Spyros Voulgaris1, Ma´rk Jelasity2, and Maarten van Steen1 1 Department Computer Science, Faculty of Sciences, Vrije Universiteit Amsterdam , The Netherlands {spyros,steen} 2 Department of Computer Science, University of Bologna, Italy Abstract. The newscast model is a general approach for communica- tion in large agent-based distributed systems. The two basic services— membership management and information dissemination—are implemented by the same epidemic-style protocol. In this paper we present the news- cast model and report on experiments using a Java implementation. The experiments involve communication in a large, wide-area cluster com- puter. By analysis of the outcome of the experiments we demonstrate that the system indeed shows the scalability and dependability proper- ties predicted by our previous theoretical and simulation results. 1 Introduction The popularity of peer-to-peer systems in the last couple of years illustrates how the Internet is gradually shifting toward a distributed system that supports more than only client-server applications. A key issue in peer-to-peer systems is that distribution of data and control across processes is symmetric. Moreover, this distribution is done in such a way that processes are highly autonomous and independent of each other. The important advantage of this approach is scalability. A well-designed peer-to-peer system can easily scale to millions of processes, each of which can join or leave whenever it pleases without seriously disrupting the system’s overall quality of service. A crucial aspect of large-scale peer-to-peer systems is that they are easy to manage. Any system that attempts to centrally manage how processes connect to each other and distribute data and control will fail, notably when processes join and leave all the time. Instead, it should be a property of the design itself that the distribution of data and contro

There are no comments yet on this publication. Be the first to share your thoughts.


Seen <100 times