A game theoretic approach to video streaming over peer-to-peer networks

https://doi.org/10.1016/j.image.2012.02.015Get rights and content

Abstract

We consider the problem of foresighted multimedia resource reciprocation in peer-to-peer (P2P) networks, which consist of rational peers aiming at maximizing their individual utilities. We introduce an artificial currency (credit) to take into account the characteristics of different parts of the video signal. The resource reciprocation with the proposed credit metric can be formulated as a stochastic game, in which the peers determine their optimal strategies using Markov Decision Process (MDP) framework. The introduced framework can be applied to the general video coding, and in particular, is suitable for the scalable video where various parts of the encoded bit stream have significantly different importance for the video quality.

Introduction

With the rapid advancement in network technologies, multimedia services, and especially video-related services, including IPTV, broadcast of real-time sports, video-on-demand (VoD), etc., have gained great popularity and hence have become a significant contributor to today's Internet traffic.

Multimedia services, such as IPTV, broadcast of sports and events, video-on-demand (VoD), and multimedia sharing have become increasingly popular and contribute significantly to today's Internet traffic. Many of these services require media streaming to a large number of subscribers. Being specifically designed for such applications, IP Multicast is the most efficient vehicle for these services. However, lack of incentives to install multicast-capable routers to handle multicast traffic, due to many political and economical issues, has prevented IP Multicast from being widely deployed. Traditional server-based solutions are often used for multimedia content delivery in such applications. However, server-based solutions are very cost inefficient and have serious scaling issues. Therefore, especially in recent years, there has been significant interest in the use of peer-to-peer (P2P) technologies for Internet multimedia services. P2P solutions are highly cost efficient since they do not require special network infrastructure support and thus are easy to deploy. Furthermore, P2P approaches scale very well with large number of subscribers since greater demand usually brings greater resources along.

These characteristics also make P2P solutions extremely attractive for file sharing and on-demand media streaming. P2P file sharing solutions such as BitTorrent [2] are extremely popular and are responsible for a large percentage of today's Internet traffic. Moreover, several multimedia streaming systems have been successfully developed for P2P networks. These systems, based on the technologies they use, can be categorized into tree-based or data-driven approaches [3], [4]. The vast majority of the existing research on P2P focuses on tree-based approaches, where peers are organized into fixed structures (i.e., trees) for delivering data. Nodes in the structure typically have well-defined relationships, e.g., “parent-child” relationships in trees. Parents are responsible for forwarding data packets they receive to their children. Consequently, the structure plays an essential role in the average user experience and thus has to be fully optimized. In addition, maintaining the tree structures has been shown to be very challenging since nodes can join and leave the group at any time. This maintenance is especially crucial for time-constrained content. In contrast to tree-based approaches, data-driven approaches do not require constructing and maintaining an explicit structure; instead, data availability is used to guide the data flow. In this work, we focus on data-driven P2P systems such as CoolStreaming [5], GnuStream [6], PeerStreaming [7] and Chainsaw [8] for multimedia streaming, or BitTorrent systems [2], [9], for general file sharing.

In these systems, data (i.e., multimedia/graphics streams or general files) are divided into smaller pieces which are distributed over the network, thus, allowing multiple people to also upload data to others while they are streaming/downloading. Nodes periodically exchange data availability information with a set of partners to notify them of the pieces they have available. While this approach has been already shown to successfully work for P2P multimedia streaming, two important aspects of this have been rarely addressed in the existing literature: (1) the optimal resource reciprocation policy for self-interested peers to maximize their individual utility and perhaps build incentives in the network for sharing more resources; (2) the role of the multimedia signal characteristics in deriving the optimal policy to enhance quality-of-service (QoS).

One of the most popular resource reciprocation strategies, the tit-for-tat (TFT) policy is deployed in the BitTorrent protocol. A peer in BitTorrent systems equally divides its available upload bandwidth among multiple leechers who contribute the most to its download [2]. However, this policy is based on the assumption of equal upload bandwidth distribution, and thus, is not optimal for heterogeneous content and diverse peers (with different upload to download ratios). An alternative resource reciprocation policy is introduced in [5] which is based on a heuristic scheduling algorithm and allows the peers to determine the suppliers of required pieces and select the peer with the highest bandwidth. A reciprocation algorithm is also proposed in [8] based on random piece selection.

The traditional assumption of the nature of the peers in a P2P network has been that the peers are altruistic and will follow the prescribed protocols without deviation (see for example [8]). However, this view has changed in many recent research works. Peers are assumed to be autonomous and rational [10] and are, thus, incentivized using economic principles [10], [11]. Consequently, in order to take into account the complex interactions of the self-interested and rational peers, game theoretic approaches have been proposed [10], [11], [12], [13]. For example, in [11] an incentive-based scheme is designed such that the benefits that a peer can draw from the system are linked to its contribution. Therefore, rational and strategic players are motivated to contribute more to the system in order to maximize their utilities. Peers' decisions in these P2P systems are considered to be myopic. A foresighted decision making solution is proposed in [14], in which, peers determine their resource distributions by explicitly considering the probabilistic behaviors of their associated peers. In [14], the resource reciprocation game is formalized as a Markov Decision Process (MDP) which enables foresighted decision making. This MDP framework is modified in [15] to take into account the limited ability of peers to characterize their resource reciprocation status. The relationship between the reciprocation complexity and the resulting utility is studied and an algorithm is developed to determine the tradeoff. In [16], a pricing mechanism is introduced to bring in incentives for sharing and hence to improve efficiency. These studies, however, do not take the multimedia characteristics, including the stringent delay constraints, into account and assumes that video quality is simply a function of the downloading rate. In addition, another drawback of such a system is its exponentially growing computational complexity that becomes unmanageable when the number of peers in the group becomes large.

In this work, we envisage a practical peer-to-peer system specifically designed for multimedia sharing. Similar to [14], peers are assumed to be self-interested, strategic players which make foresighted decisions in order to maximize their own video quality. An MDP is employed to obtain the optimal resource allocation policy based on the other peers' reciprocation history. However, an approximation method is introduced in order to reduce the computational burden of the optimization. Since different pieces of the bit stream may have significantly different impact on the video quality, a fictitious currency is introduced to allow peers to offer different prices for various parts of their multimedia content [1]. Furthermore, at the packet level, the problem of optimal packet and packet-request scheduling is addressed.

This paper is structured as follows. In Section 2 we provide a brief introduction to multimedia distribution systems in P2P networks. Section 3 proposes the content-aware resource measurement and discusses the fictitious currency used to quantify the importance of multimedia content. In Section 4 we elaborate on the resource reciprocation framework in detail and introduce the solution algorithm. Experimental results are presented in Section 5, and conclusions are drawn in Section 6.

Section snippets

Multimedia distribution in P2P networks

In order to efficiently address the high variability in demand for video quality and resources each peer contributes to the network, scalable video coding (SVC) has been used in this work. However, our formulation is general and any video coding scheme may be used. A video bit stream is called scalable when parts of it can be removed in a way that the resulting substream forms a valid bit stream representing the content of the original with lower resolution and/or quality. SVC is a video

Content-aware resource measurement

Since various blocks have different impact on the quality of the reconstructed video, an incentive must be implemented within the system to encourage peers to prioritize and allocate more resources to the more important blocks. In order to provide that incentive, we introduce a fictitious currency called “credit”. Depending on the requesting block's PID, a peer i offers ρ[1,+) credits per bit to peer j who replies to its request for the missing block. This agreed upon price is included in the

Resource reciprocation game

Resource reciprocation process in P2P networks can be modeled as a game environment in which the peers (players) are interested in obtaining multimedia contents from each other [14]. From the perspective of a single peer, the resource reciprocation game involves itself and a group of other peers associated with it. Examples of such a group include swarms [9], [18], partnerships [5] and neighbors [8]. The group of peers associated with peer i is denoted by Gi, which does not include peer i

Experimental results

In this section we present the results of our simulations to show the effectiveness of the proposed techniques. We assume that a group consisting of n nodes has already been formed. Before the streaming starts, nodes send out a BM message notifying each other about their available content. In our simulations, the size of the notification messages are ignored in the network since they are usually much smaller than data packets. However, in a real network simulation the effect of these messages

Conclusion

In this paper we considered the problem of content-aware, foresighted resource reciprocation for multimedia content streaming over P2P networks consisting of self-interested and rational peers. We have taken the video content and signal characteristics into account by proposing an artificial credit currency. The resource reciprocation process was modeled as a stochastic model and an MDP framework was introduced to determine the optimal strategies. The performance improvement by the proposed

References (23)

  • E. Maani, A.K. Katsaggelos, A game theoretic approach to video streaming over peer-to-peer networks, in: IEEE 17th...
  • BitTorrent,...
  • J. Liu et al.

    Opportunities and challenges of peer-to-peer Internet video broadcast

  • S. Androutsellis-Theotokis et al.

    A survey of peer-to-peer content distribution technologies

    ACM Computing Surveys

    (2004)
  • X. Zhang et al.

    CoolStreaming/DONet: a data-driven overlay network for peer-to-peer live media streaming

  • X. Jiang et al.

    GnuStream: a P2P media streaming system prototype

  • J. Li, PeerStreaming: A practical receiver-driven peer-to-peer media streaming system, Microsoft Research...
  • V. Pai, K. Kumar, K. Tamilmani, V. Sambamurthy, A.E. Mohr, Chainsaw: eliminating trees from overlay multicast, in:...
  • A. Legout et al.

    Clustering and sharing incentives in BitTorrent systems

  • B. Yu, M.P. Singh, Incentive mechanisms for peer-to-peer systems, in: Agents and Peer-to-Peer Computing Second...
  • C. Buragohain et al.

    A game theoretic framework for incentives in P2P systems

  • Cited by (8)

    • Network aware peer-to-peer media streaming: Capacity or proximity?

      2014, Computer Networks
      Citation Excerpt :

      Ma et al. propose a resource distribution mechanism so as to provide service differentiation based on each peer’s contribution levels in resource sharing. Maani et al. model the resource reciprocation among peers as a Markov Decision Process (MDP) [34], based on which peers determine their optimal strategies for cooperation. Asioli et al. [35] devise an optimal resource allocation scheme by combining minimum delay functionalities with an incentive provision mechanism.

    • Client-server games and their equilibria in peer-to-peer networks

      2014, Computer Networks
      Citation Excerpt :

      In [8] the authors study unilateral or bilateral network formation games, where peers individually open up connections on multiple paths or mutually agree on setting up paths respectively. Ref. [9] addresses the problem of optimal multimedia P2P content exchange among a set of peers. The emphasis here is on a resource reciprocation game.

    • A reliable peer-to-peer streaming protocol in low-capacity networks

      2021, Peer-to-Peer Networking and Applications
    View all citing articles on Scopus

    An earlier version of this paper was presented at the 2010 IEEE International Conference on Image Processing (ICIP), Hong Kong [1].

    View full text