Coercion builds cooperation in dynamic and heterogeneous P2P live streaming networks
Introduction
Incentive provision [1] is a critical factor for the success of a peer-to-peer (P2P) network that relies heavily on peers’ cooperation to sustain reasonable system performance. Indeed, in the past decade, various effective schemes have been proposed and implemented in practical systems [2], [3], [4], [5], [6], [7]. However, such schemes, typically based on reciprocity in the form of bilateral data exchange (i.e., tit-for-tat like behaviors) or virtual payments [1], are useful mostly for file sharing and video-on-demand (VoD) systems, but not for the increasingly popular live video streaming systems, e.g., PPLive [8], PPStream [9] and UUSee [10], serving millions of active daily users spread across the globe [11]. The key problem is that with stringent playback deadlines for live streaming, there is a lack of opportunity for timely data exchange.
Specifically, Piatek et al. drew on real measurements of tens of thousands of peers in PPLive [8] and demonstrated the existence of limited trading opportunities among neighboring peers that decisively invalidated the efficiency of bilateral data chunk exchange mechanisms for P2P live streaming [12]. The key insight is that with the stringent time constraint, a more effective strategy is to support more judicious peer selection, so as to pair up peers to allow for timely exchange of relevant data chunks. Indeed, peer selection is critical to the effectiveness of the network topology in terms of connectivity, and hence, system performance, by establishing cooperative local connections leading to a high performance global overlay topology.
Broadly speaking, there are two different kinds of efficient peer selection schemes: capacity aware [13], [12] and locality aware [14], [15]. The essence of the former is to place high capacity peers closer to the source server so as to better utilize their upload capacity and achieve optimal system streaming quality. On the other hand, the latter promotes network friendliness by matching the overlay topology with the underlying physical networks. For instance, this can be achieved by clustering peers from the same ASes and thus reducing financial cost incurred by inter-AS traffic. For clarity, we focus our analysis on capacity aware peer selection that can be easily extended to locality aware peer selection. The efficiency of our proposed strategies is evaluated in both peer selection scenarios.
As a pioneering effort, Piatek et al. proposed Contracts [12] to provide incentives against free-riding through capacity aware peer selection. To the best of our knowledge, Contracts is the only study to provide practical incentives by structuring topologies via judicious peer selection, instead of bilateral chunk exchanges [12]. Unfortunately, the success of Contracts is contingent on the presumed cooperation of peers in the peer selection process. To further aggravate the problem, it is as yet a difficult challenge to tackle the non-cooperative issue in a dynamic and heterogeneous network (i.e., peers are equipped with diverse upload/download bandwidth). With this motivation, this paper aims to propose practical incentive schemes for P2P live streaming, by fully exploring strategic peer selection.
Our contributions in this paper are as follows. We first demonstrate quantitatively that peer non-cooperation is highly likely in strategic peer selection. We then model strategic peer selection as a repeated game, based on which we propose a Striker strategy for coercing peers to cooperate in P2P live streaming systems. Our Striker strategy works well in both centralized and fully distributed network environments. The key idea of the Striker strategy is to provide enough threat to selfish peers to thwart them from deviating from cooperation. The most important feature of our scheme is that it provides peers with no incentive to falsify their upload capacities. We are also among the first to study quantitatively the hidden effects of ineffective incentive protocols (i.e., crowding out of low bandwidth peers).
Contrary to previous studies [13], [12], both our empirical simulations and rigorous analytical modeling show that, due to the inevitable peer non-cooperation, prescribed protocols induced by capacity aware peer selection significantly deteriorate system social welfare, let alone provide sharing incentives in Nash equilibria (N.E.). In particular, such performance degradation is incurred by clustering among peers with similar upload capacities that is different from file sharing and VoD systems [16], [17], due to different system requirements, such as stringent playback deadlines.
The remainder of this paper is organized as follows. In Section 2, we present the system model of the interactions between the content provider and selfish peers, and a solution concept to quantify any peer selection protocol. Extensive simulation results show that noncooperation of peers in strategic peer selection is a general problem not limited to capacity aware peer selection. Formal game theoretic analysis is then provided to model iterative peer selection in Section 3. Section 4 presents our Striker strategy for coercing selfish peers to cooperate in establishing effective topologies, followed by practical algorithms in Section 5. Section 6 validates the efficiency of our Striker strategy with extensive simulations. Recent advances in P2P networks are surveyed in detail in Section 7. Finally, we present conclusions and future work in Section 8.
Section snippets
System architecture and model
In this section, we first present the system architecture and model. Before delving into our detailed rationality analysis and incentive scheme design, we quantitatively scrutinize implications of strategic peer selection for P2P live video streaming based on extensive simulation results, augmented with preliminary modeling efforts. Indeed, such preliminary analysis shows that existing overlay construction protocols based on capacity aware peer selection deteriorate social welfare in Nash
Game theoretic analysis
In this section, based on the above empirical observations, we mathematically formulate the game model of interactions among trackers and peers, followed by the modeling of interacting heterogeneous peers. Under the scenario of multiple bandwidth classes, we formally show that cooperation of high capacity peers is not a Nash equilibrium.
Coercing non-cooperative peers
In this section, repeated games for iterative peer selection demonstrate the feasibility of coercion leading to cooperation. Then, we propose a novel Striker strategy to coerce high capacity peers to cooperate. The crux is to provide enough threat to non-cooperative behaviors of high capacity peers so as to prevent strategic deviation from cooperation. We first constrain that peers can only obtain neighbor lists via trackers. We then relax this constraint and demonstrate that our incentive
Algorithms to stimulate cooperation
Theoretically, peers and trackers can adjust parameters such as their cooperative probability, threshold for cooperation detection, and punishment duration by maximizing their utility. However, the lack of close-form functional relationships between these parameters and the utility increases the problem complexity.
Let us present several simple yet effective algorithms to implement the Striker strategy in P2P live video streaming networks. As discussed above, there are two general approaches to
Evaluation of proposed algorithms
We conduct extensive simulations to evaluate our models and the Striker strategy. First, we describe our simulation setting for P2P live video streaming. Then, we present and discuss our simulation results.
Related work
Game theory has been widely used for incentive analysis. Buragohain et al. [2] performed the pioneering research of utilizing game theory to model and analyze interactions among rational and strategic peers as a non-cooperative game. However, the model and analysis was too general. Zhao et al. [4] provided a concrete mathematical framework by stochastically investigating incentive policies. Yeung and Kwok [39] designed a repeated packet exchange game to motivate reciprocal chunk sharing between
Concluding remarks
In this paper, we provide practical incentives for P2P live streaming, by fully exploring strategic peer selection. A strategic client judiciously adjusts its neighbor set by varying the probability of cooperation. Extensive simulations reveal hidden effects of strategic peer selection. In particular, non-cooperation of high capacity peers crowds out low capacity peers. Tractable game models are formulated to analyze interactions among trackers and heterogeneous peers that formally demonstrate
Xin Jin received his B.Eng. degree in communication engineering from University of Electronic Science and Technology of China, Chengdu, China, in 2008. He received his Ph.D. degree in Electrical and Electronic Engineering from the University of Hong Kong in 2013. From September 2013 to August 2014. He was a Postdoctoral Research Fellow in the Pennsylvania State University. He is now a Software Developer Engineer in Yahoo Inc., Sunnyvale, CA, United States. His main research interests are
References (41)
Peer-to-Peer Computing: Applications, Architecture, Protocols, and Challenges
(2011)- C. Buragohain, D. Agrawal, S. Suri, A game theoretic framework for incentives in P2P systems, in: Proc. of IEEE P2P,...
- et al.
On game theoretic peer selection for resilient peer-to-peer media streaming
IEEE Trans. Parall. Distr. Syst.
(2009) - B.Q. Zhao, J.C. Lui, D.-M. Chi, Analysis of adaptive incentive protocols for P2P networks, in: Proc. of IEEE INFOCOM,...
- et al.
Incentive cooperation strategies for peer-to-peer live multimedia streaming social networks
IEEE Trans. Multimedia
(2009) - W. Wu, J.C. Lui, R.T. Ma, Incentivizing upload capacity in P2P-VoD systems: a game theoretic analysis, in: Proc. of...
- et al.
Free-riding and whitewashing in peer-to-peer systems
IEEE J. Sel. Areas Commun.
(2006) - PPLive, 2013....
- PPStream, 2013....
- UUSee, 2013....
A measurement study of a large-scale P2P IPTV system
IEEE Trans. Multimedia
Location awareness in unstructured peer-to-peer systems
IEEE Trans. Parall. Distr. Syst.
Dynamic bandwidth auctions in multioverlay P2P streaming with network coding
IEEE Trans. Parall. Distr. Syst.
Building heterogeneous peer-to-peer networks: protocol and analysis
IEEE/ACM Trans. Network.
Cited by (5)
Incentive Mechanisms in Peer-to-Peer Networks - A Systematic Literature Review
2023, ACM Computing SurveysContent-based load balancing of tasks using task clustering for cost optimisation in cloud computing environment
2022, International Journal of Advanced Intelligence ParadigmsOverlay Convergence Analysis in P2P Networks: An Assessment of the 2PC Algorithm
2020, 2020 International Conference on Innovation and Intelligence for Informatics, Computing and Technologies, 3ICT 2020Optimality analysis of locality-aware tit-for-tat-based P2P file distribution
2020, Peer-to-Peer Networking and ApplicationsModeling of free riders in P2P live streaming systems
2017, 2017 International Conference on Computing, Networking and Communications, ICNC 2017
Xin Jin received his B.Eng. degree in communication engineering from University of Electronic Science and Technology of China, Chengdu, China, in 2008. He received his Ph.D. degree in Electrical and Electronic Engineering from the University of Hong Kong in 2013. From September 2013 to August 2014. He was a Postdoctoral Research Fellow in the Pennsylvania State University. He is now a Software Developer Engineer in Yahoo Inc., Sunnyvale, CA, United States. His main research interests are incentive provision and performance modeling of distributed systems including P2P networks and cloud computing systems.
Y.-K. Kwok (Ricky) received a B.Sc. degree in Computer Engineering from the University of Hong Kong in 1991, and the M.Phil. and Ph.D. degrees, both in Computer Science, from the Hong Kong University of Science and Technology (HKUST) in 1994 and 1997, respectively. Before returning to the University of Hong Kong as an Assistant Professor in August 1998, he was a Visiting Scholar for one year in the Parallel Processing Laboratory, the School of Electrical and Computer Engineering at Purdue University (West Lafayette, Indiana, USA), where he worked on a DARPA research project in the field of distributed heterogeneous computing. During his sabbatical leave year (from August 2004 to July 2005). He worked as a Visiting Associate Professor in the Internet and Grid Computing Laboratory, Department of Electrical Engineering–Systems at the University of Southern California (Los Angeles, California, USA), where he conducted research on several interesting Internet security problems such as DDoS defense and traceback, worm containment and signature generation, etc. From July 2007 to May 2009, he also worked as an Associate Professor in the Department of Electrical and Computer Engineering at the Colorado State University (Fort Collins, Colorado, USA), where he taught courses and conducted research in the areas of wireless sensor networks, high performance computer architecture, reconfigurable computing systems, and heterogeneous computing.
His research focus has been on designing efficient communication protocols and robust resources management algorithms toward enabling large scale distributed mobile computing. In these research areas, he has authored one textbook, co-authored another two textbooks, and published more than 200 technical papers in various leading journals, research books, and refereed international conference proceedings. He is a Senior member of the Association for Computing Machinery (ACM). Ricky is a Fellow of the HKIE, the IEEE, and the IET. He is also currently serving as Associate Vice-President (Teaching and Learning), overseeing various e-learning initiatives at HKU. He is also a member of the IEEE Computer Society and the IEEE Communications Society. He is an Associate Editor for the IEEE Transactions on Parallel and Distributed Systems. From March 2006 to December 2011. He served on the Editorial Board of the Journal of Parallel and Distributed Computing as a Subject Area Editor in Peer-to-Peer Computing.
He received the Outstanding Young Researcher Award from the University of Hong Kong in 2004. In January 2010, one of his journal papers was ranked #4 among top ten all time most cited papers published in the IEEE Transactions on Parallel and Distributed Systems, based on Scopus and Google Scholar citation counts as of October 2009.