Performance Evaluation of Distributed Prefetching for Asynchronous Multicast in P2P Networks ⋆ Abhishek Sharma 1 , Azer Bestavros 2 , and Ibrahim Matta 2 1 Elec. & Comp. Eng., Boston University, Boston, MA 02215, USA abhishek@bu.edu 2 Comp. Sc., Boston University, Boston, MA 02215, USA {best,matta}@cs.bu.edu Abstract. We consider the problem of delivering real-time, near real- time and stored streaming media to a large number of asynchronous clients. This problem has been studied in the context of asynchronous multicast and peer-to-peer content distribution. In this paper we eval- uate through extensive simulations the performance of the distributed prefetching protocol, dPAM [20], proposed for scalable, asynchronous multicast in P2P systems. We show that the prefetch-and-relay strategy of dPAM can reduce the server bandwidth requirement quite signifi- cantly, compared to the previously proposed cache-and-relay strategy, even when the group of clients downloading a stream changes quite fre- quently due to client departures. 1 Introduction On-demand media distribution is fast becoming an ubiquitous service deployed over the Internet. The long duration and high bandwidth requirements of stream- ing media delivery present a formidable strain on server and network capacity. Hence, scalable delivery techniques, both in terms of network link cost as well as server bandwidth requirement, are critical for the distribution for highly popular media objects. For the delivery of real-time media to synchronous requests, multicast solu- tions (whether using network support in case of IP multicast or using end-system support through peer-to-peer networks) are attractive as they reduce both net- work link costs and server bandwidth requirements for serving a large number of clients [22,5,13,11]. However, a number of scenarios can be envisioned in which the client requests for streaming media objects are likely to be asynchronous. This is true for requests to stored streaming media objects (e.g., on-demand delivery of popular movie clips or news briefs to clients), as well as for requests to buffered live streams (e.g., playout of a webcast to a large number of clients requesting that webcast asynchronously but within a short interval). To enable asynchronous access to streaming media objects, various IP mul- ticast based periodic broadcasting and stream merging techniques [22,13,11,17] ⋆ This work was supported in part by NSF grants ANI-9986397, ANI-0095988, EIA- 0202067 and ITR ANI-0205294. C.-H. Chi, M. van Steen, and C. Wills (Eds.): WCW 2004, LNCS 3293, pp. 154–168, 2004. c  Springer-Verlag Berlin Heidelberg 2004