ON THE DESIGN OF PREFETCHING STRATEGIES IN A PEER-DRIVEN VIDEO ON-DEMAND SYSTEM Yanming Shen†, Zhengye Liu†, Shivendra Panwar†, Keith Ross‡, Yao Wang† Department of Electrical and Computer Engineering† Department of Computer and Information Science‡ Polytechnic University ABSTRACT In this paper, we examine the prefetching strategies in a peer- driven video on-demand system. In our design, each video is encoded into multiple low bit-rate substreams and copies of the substreams are distributed to the participating peers. When a peer streams in a substream of rate r, it instead streams at rate ˆ r where ˆ r>r. In this manner, if one of the peer’s suppliers disconnects, the client peer can tap the reservoir of prefetched bits while searching for a replacement server, thereby avoiding any glitches or reduced visual quality. We examine how to assign prefetching rates to each of substreams as a function of their importance. Our studies show that ap- propriate prefetching strategies can bring significant perfor- mance improvements for both multiple description and lay- ered videos. 1. INTRODUCTION We have proposed a peer-driven video on-demand architec- ture [1, 2]. In our design, as shown in Figure 1, each video is encoded into multiple low bit-rate substreams, and copies of these substreams are stored in peers. When a client wants to view a video, it receives multiple sub-streams, each from a different server peer. As the sub-streams arrive to the client, the client combines the sub-streams, decodes and displays the video. Because each sub-stream typically has a rate that is a fraction of the combined stream, the server peers can more easily accommodate sub-streams with their limited upstream bandwidth. Furthermore, if the system is designed properly, the loss of one stream due to a server failure or disconnect will not seriously impair video quality while it is waiting for a replacement sub-stream. In this paper, we investigate how to design prefetching policies that improve the system performance. By prefetch- ing, peers download streams at rates that are higher than the playback rate [3, 4, 5, 6]. In this manner, when a substream is lost, the client may have a sufficient “reservoir” for that sub-stream, so that playback continues without any quality degradation. This paper is organized as follows. In Section 2, we describe our model and in Section 3 we present both S 1 S Client S 2 S M Substream 1 Substream 2 Substream M . . . Request Fig. 1. Peer-driven video on-demand architecture (A client sends the request to the video server, and the server finds M peer servers which store substreams of the video. Then, each peer server sends a different substream to the client.) an optimal and a heuristic prefetching policy. We evaluate the performance of our proposed scheme with simulation in Section 4, and Section 5 concludes the paper. 2. SYSTEM MODEL We consider a homogeneous system with N peers, each with B u bps of uplink bandwidth. Each peer is connected with probability µ and peer connectivity is independent from peer to peer. There are J videos in the network. Each video is encoded into M substreams using either multiple description coding (MDC) or layered coding [1, 2], and each substream has a bit rate of r. Thus the total rate of a video with all substreams is R = Mr. Each peer has a storage constraint and stores at most one substream of a particular video. When a user makes a request, the peers storing substreams of that video are selected and each of these server peers sends a different substream to the client. When the available uplink bandwidth of a server peer exceeds the substream rate, the system is prefetched into the client’s prefetch buffer, which we model as infinite. This allows the peer to stream at rate ˆ r>r and build up a reservoir of non-rendered video. In this manner, if one of the server peers disconnects, the client peer can tap into the reservoir while searching for a replacement peer, thereby avoiding any glitches or reduced visual quality. By increasing ˆ r, we build up the reservoir more quickly, 817 1424403677/06/$20.00 ©2006 IEEE ICME 2006