An effective mesh-pull-based P2P video streaming system using Fountain codes with variable symbol sizes

Abstract

In this paper, we propose an effective mesh-pull-based video streaming system over P2P networks using Fountain codes with variable symbol sizes for video-on-demand services. The goal of the proposed system is to provide a stable video streaming service of high quality with minimum computational complexity and a short initial latency over P2P networks. Basically, Fountain codes are adopted in the proposed system to simplify the handshaking procedure, which causes a large initial latency, and to support a robust video streaming service despite packet losses. The proposed Fountain encoding mechanism works by using feedback information to reduce unnecessary encoded symbol generation. In addition, the Fountain code symbol size is continuously adjusted to minimize additional computational overhead required for Fountain encoding/decoding. The proposed system is fully implemented in software and examined over an Internet environment.

Introduction

In recent years, P2P (Peer-to-Peer) based networking technology has attracted a huge amount of research interest and has now become the dominant content distribution method over the Internet. It operates with minimal (or no) reliance on always-on infrastructure servers. Many P2P applications [1], [2] have already been deployed over the Internet, such as Gnutella [3], PPLive [4], Skype [5], and BitTorrent [6]. Until now, file-sharing services have been a dominant P2P application using networking technology, but in recent years video streaming services have quickly emerged over P2P networks. In general, video streaming services require strict QoS (Quality of Service) including bandwidth, delay, and jitter compared to traditional data services.

In general, P2P-based networking technology can be classified into two groups: tree-based techniques and mesh-based techniques. In tree-based techniques, a peer has a parent peer and multiple child peers; a peer receives data from a parent peer and relays the received data to child peers. Yoid [7], ALMI [8] and HMTP [9] are the tree-based techniques. Tree-based techniques have a short delay since a video stream is basically pushed along well-defined routes. However, when a peer leaves or packet loss occurs, all of the descendants experience QoS degradation. To overcome the drawbacks of tree-based techniques, mesh-based techniques have been proposed. In mesh-based techniques, a peer has multiple links to other peers. There are no parent and child relationships; peers send and receive streaming data to each other and pull delivery is adopted as the delivery mechanism. CoolStreaming [10], Bullet [11], Chainsaw [12], and IPTV over P2P networks [13] are examples of mesh-based techniques. The mesh-pull-based P2P video streaming system generally uses a buffer map and a sliding window to support QoS requirements for video streaming service. A buffer map indicates whether or not chunks are cached at the storage devices of a peer, and a sliding window supports seamless playback by receiving chunks only in the sliding window. Fig. 1 shows the basic concept of the traditional P2P video streaming based on a buffer map and sliding window. The transmission procedure is the following. If a peer wants a video streaming service, it sends a request message to other peers. Peers that have the requested video stream transmit the buffer map information to the receiver-peer. The receiver-peer makes a schedule for the chunks in the sliding window considering the network environment. Generally, this procedure requires huge overhead and a long initial latency since it has to be exchanged periodically.

In general, mesh-based techniques are resilient against the leave/failure of peers and packet loss. However, they have a long delay and more control overhead for data transmission. So far, some P2P-based multicast tree construction algorithms have been proposed to minimize the (average/maximum) delay by considering only the link delay values [7], [8], [9], [10], [11], [12], [13], [14], [15]. These hybrid-based techniques, integrating tree-based and mesh-based overlays techniques, aim to improve both initial latency and resilient [16], [17]. Recently, several networking coding schemes [18], [19], [20] have been proposed to effectively improve the throughput and initial latency. Avalanche [21], [22] and Lava [23] show that network coding may improve the overall performance of P2P networking. Fountain codes can be a solution to enhance resilience, improve network utilization, and reduce control message overhead over P2P networks [24], [25], [26]. Thomos and Frossard proposed a network coding algorithm based on Raptor codes for video streaming services [27]. R² [28] is designed to improve the performance of live streaming in terms of initial buffering delay, resilience to peer uncertainty, and server capability. Grangetto et al. proposed a Fountain code based video streaming system [29] to reduce initial delay and simplify the protocol design over P2P networks.

In this paper, we propose an effective mesh-pull-based P2P video streaming system using Fountain codes with low initial latency for video-on-demand services. This system dynamically controls the symbol size of Fountains codes to minimize computational complexity. This is one of the unique features of the proposed algorithm. The remainder of this paper is organized as follows. A brief review of fountain codes is presented in Section 2, the details of the proposed mesh-pull-based P2P video streaming system in Section 3, experimental results in Section 4, and concluding remarks in Section 5.