THE JOURNAL OF CHINA UNIVERSITIES OF POSTS AND TELECOMMUNICATIONS Volume 13. Issue 4, December 2006
ZHANG Jian-feng, WANG Ru-chuan
An agent-based incentive mechanism for P2P systems CLC number TP393.01
Document A
Abstract With the increasing uses of peer-to-peer (P2P) systems, the problems of fair information and resources sharing become serious. P2P systems are self-organizing, distributed systems with no centralized authority, such as Free-riding and Tragedy of Commons. Because of the voluntary participation and lack of motivation, the information and resources available in P2P networks are extremely variable and unpredictable. This article studied the problems existing in P2P networks and propose a different method to stimulate the peers in P2P networks to share information and resources, using agents so as to improve the performance of P2P networks. And through simulation experiments and analyzing the results, it can be concluded that this mechanism can effectively solve the problems of fair sharing in P2P systems .
Keywords P2P,fair, sharing, motivate, agent
1 lntroductlon P2P systems differ from the traditional distributed computing systems. They are self-organizing, distributed resource-sharing networks. In P2P systems, there is no central authority that controls or manages the various components. Nodes can join or exit from P2P systems without any control. By pooling together the resources of various autonomous machines, P2P systems are able to provide an inexpensive platform for distributed computing, storage, or data-sharing that is highly scalable, available, fault tolerant, and robust [I].With the rapid growth of decentralized and structured or unstructured P2P networks, several commercial P2P systems, such as Napster, and Gnutella have emerged [2, 31. However, there exist some problems in the P2P systems because of the rational nodes, which do not contribute information or resources to the system. Free-riding and the Tragedy of Commons are the two major problems in the P2P systems. As is reported, nearly 70% of Gnutella users do not share any file with other in a P2P Received date: 2006-02-04 ZHANG Jian-feng, WANG Ru-chum (r?) College of Computer, Nanjing University of Posts and Telecommunications, Nanjing 210003, China WANG Ru-chuan State Key Laboratory for Novel SoftwareTechnology, Nanjing Universtiy, Nanjing 210093, China E-mail:
[email protected]
Article ID
1005-8885(2006)04-0081-05
community and nearly 50% of all search responses come from the top 1% of content sharing nodes [4].Selfish peers who do not contribute to a group are known as free riders (Atar and Huberman, 2000). The Free-riding problem is not simply theoretical. To eliminate this problem, practically useful mechanisms are needed to encourage peers to contribute service (Golle, et al., 2001). Nodes, which share information and resources, are prone to congestion, and this leads to the Tragedy of Commons [5].During the development of Internet, Tragedy of Commons is also a serious problem [6, 71. There are also other problems, for example, many users intentionally misrepresent their connection speeds so as to discourage others from going to their nodes for downloading the file. Still worse, the P2P systems give no service differentiation between users who do not share any information with others or make any contribution to the P2P community. In this article, we propose a different mode of incentive mechanism for P2P systems, which is an agent-based incentive mechanism. An agent can be described as a software component that performs a specific task autonomously on behalf of a person or organization [8].
2 PZPnetworks A P2P network is a novel mode of communications, and it comprises a set of nodes that can cooperate and collaborate each other in a decentralized and distributed manner. There is no central authority to mandate or coordinate the resources that are contributed by the peers [l].The P2P technology can be used in several domains, such as files exchange, distributed computing, distributed search and so on. Unlike the traditional clienthewer structure, each node in a P2P system plays the role of a client as well as server. P2P networks offer high availability, better bandwidth through many users' wideband connections, and better scalability without central servers as bottlenecks. The impact of P2P in file sharing is significant. Hundreds of millions of copies are already in use, and this makes file sharing as the most popular application on the World Wide Web. Therefore, P2P networks create a huge content sharing base and a complex value network, which offers novel business opportunities and models to various actors, such as network operators, and also, to novel actors like
82
The Journal of CHUPT
multimedia producing communities. In this article, we mainly study the incentives for files exchange. In fact, the P2P technology originated from files exchange technology, and the evolvement of files exchange technology is the most significant step in the development of P2P networks. 2.1 Nodes In P2P networks Compared to traditional computing systems, the major difference of a P2P network is its social economy complexity. The whole system is composed of thousands of nodes, which have their own functions, so the action of each node is unpredictable because of the different functions of individual nodes. The functions of these nodes have different traits. We can divide these nodes into four types. 1) Rational nodes: The major objectives of P2P system of nodes of this type are to meet their own interests to a large extent. They will give preference to their own interests when there are conflicts between their own nodes and the system nodes. To meet their own needs, they design their own strategies using the information obtained from the system. 2) Obedient nodes: Nodes of this type strictly obey the algorithms and agreements defined in advance. 3 ) Adversary nodes: nodes of this type destroy the system with malice and make their strategies in this goal. 4) Faulty nodes: Nodes of this type cannot run in gear because of some faults. The nodes in a P2P system belong to different organizations and individuals with different interests, and effective cooperation and responsibilities mechanism is lacking, so that the nodes can just think of their own interests without considering the interest of the system. And this causes the problems mentioned above . In this article, we assume that most nodes in a P2P system are rational, which is the focus of our study.
2006
incentives mechanism for P2P networks, the different nodes share information and resources with other nodes, and thereby increase the efficacy of the whole system . From the example of Gnutella and other established P2P systems, we conclude that incentives mechanism is significant in large-scale P2P systems and complicated structure. There are several literatures that suggest incentive mechanisms for solving the resources sharing problems in p2P systems [9-121.
8 An agent-based lneentlve mechanlsm In this section, we propose an agent-based incentive mechanism for P2P systems. We assume that the incentive mechanism for a P2P system can provide a running circumstance for agents. 3.1 Agent technology
An agent is a software entity that is situated in some environment and can sense and react to the changes in that environment. Agents are capable of operating autonomously and in a goal directed manner to meet its design objectives. In a multiagent system, tasks are carried out by interacting the agents that can cooperate with each other to achieve common or private goal [ 131. An agent can identify what it needs to do to satisfy its design objectives, rather than being explicitly told what to do at any given moment. A multiagent system is defined as a system that consists of a number of agents, which interact with one another, typically by exchanging messages through some computer network infrastructure. In general case, the agents in a multiagent system will be representing or acting on behalf of users or owners with different goals and motivations. To interact effectively, these agents should cooperate, coordinate, and negotiate with each other, exactly in the same way that humans cooperate, coordinate, and negotiate in everyday lives [14, 151.
2.2 Goals of Incentives for P2P networks
3.2 With greater use of P2P networks, the problems are more serious caused by the lack of incentives, and it does badly to the development of P2P networks. The typical example is that of Gnutella. The P2P file sharing application has been plagued by the free-riding problem that most of the peers download resources from other peers without sharing its resources or contributing to the P2P system. The objective of adopting incentive mechanisms for P2P networks is to stimulate the cooperation between selfinterested participants. Some researchers found that simple incentive mechanism can be used effectively to share more resources (Ranganathan, et al., 2003). With the use of
Roles of Agent In the mechanism
We have identified several types of agents in our incentive mechanism as follows: 1 ) Download agents (DAs): Download agents are those agents that supervise, compute, and record the download fluxes of each peer. We define that a peer will score a negative mark by downloading certain bytes. The calculation of mark is defined as follows: Md = - D , / a where Md denotes the download marks that a peer scores, Db denotes the bytes that a peer downloads and a is a data that can be taken first ( aE R '). And each peer's download agent will
No. 4
ZHANG Jian-feng, et al.: An agent-based incentive mechanism for P2P systems
83
send its download mark to its charge agent when a download is accomplished. 2) Upload agents (UAs): Upload agents are those agents that supervise, compute, and record the upload fluxes of each peer. And upload agents can remember an upload resource’s name and the number of times it is downloaded with the different IP they record. In other words, if a file is downloaded by the same peer, the number of times it is downloaded will not increase. We define that a peer will score a plus point by uploading certain bytes. The mark is calculated as follows: M u =U , / a + ( n - l ) p Where Mu denotes the upload marks that a peer scores, Ub denotes the bytes that a peer uploads, a is the same to the one mentioned above, n is the times of a file being downloaded, and /3 is a data we have defined previously. We denote p as an encouraging mark gene ( p E R ’). To encourage the peers to provide resources that are more useful, we define that if more than one peer downloads the same resource, the upload peer will gain added marks as (n - 1) D . And each peer’s upload agent will send its upload mark to its charge agent when an upload is accomplished. 3) Charge agents (CAs): Charge agents are those agents that control the performance of a node’s download and decide whether a node can download resources from other peers in the P2P network by the sum marks of its download mark and upload mark. The calculation is as follows: M,=Md + Mu Here Mt denotes the total marks that a peer scores.
s,= SnMa where S, stands for the speed of a peer download at the t time, and S, stands for the normal speed of download. M,, is the absolute value of M,. This can solve the coordinated cheating problem between two peers because the speed of the download peer will be very slow when it gains more negative marks. However, in this case, the upload of the peer is in gear to make it convenient for the peer to do more contribution to the P2P system. The structure of the mechanism is shown in Fig. 1.
3.3 The main idea and archltecture of the mechanism
4 Results of the Stlmulatlon experiment
The main objective of the agent-based incentive mechanism is to identify the three types of agents we used above as mediators of the peers. The agents are part of the P2P system and the users cannot do anything to them. When a node joins into the P2P system, they will be executed. And when a node first joins into the P2P system, its charge agent has an initial mark-Mi ( M ~ ER ’). In fact, a peer cannot find other peers without executing the agents. The communications between the two or more peers in P2P networks are controlled by the agents. The charge agent of a peer updates its mark information whenever it receives mark information from the download agent or from the upload agent. We divide these cases into two types on the basis of the total mark of the charge agent as follows. First, when the total mark is plus (include 0), we call this instance plus case. Second, when the total mark is less than 0, we call this instance as negative case. Then we recommend that if a peer is in a plus case, it can share information and resources in gear. If a peer is in a negative case, the charge agents will e decrease the speed of its download as a penalty to the peer. The speed will get slower as the calculation defined as follows:
To evaluate the agent-based incentive mechanism in our simulation system we consider two conditions: a P2P system with an agent-based incentive mechanism and another P2P system with no incentive mechanism. And we assume three types of nodes are present in the two types of P2P systems. These are self-giving nodes (we call nodes of this kind SN for short), no contributing and consuming nodes (we call EN nodes of this kind for short), and rational (selfish) nodes (we call nodes of this kind RN for short). We hypothesize the initial conditions, experimental time, the parameters, and the numbers of nodes in Table 1 as follows:
Peer B
DA
UA
UA
CA
DA CA
Files for
sharing
L (CA: Charge agent; UA: Upload agent; DA: Download agent)
Fig. 1 The architecture of the incentive mechanism
Table 1 The parameters of the simulation SN/% EN/% RN/%
Simulation
MI
a
P
timesh
Number of
nodes Incentive
LO
60
30
lo0
Non-
10
60
30
100
50
100
5
50
50
incentive
In the simulation experiment, we propose thar if Mt of a node
The Journal of CHUPT
84
is less than 50, it will change to the type of RN, and if the Mt of a node is more than 50, it will change to the type of SN. Here, we present the simulation results are capable of testifying the agent-based incentive mechanism. In F'2P networks, peers can be seen as strategic entities with the goal of maximizing its network utility. There are contradictions between the individual's rationality and the social rationality. In a P2P system with no incentive mechanism, the rational nodes try to meet their own interests to a large extent, so they will not contribute to the system. While in a P2P system with the agent-based incentive mechanism, the rational nodes contribute more to the system to achieve good performance. To testify the agent-based incentive mechanism, we mainly concern the changes of rational nodes in the simulation experiment. The results of the two types of conditions are shown in Fig. 2. From the curves of the two types of conditions shown in Fig. 2, we can see that rational nodes in the system are very less with the agent-based incentive mechanism. Finally, we can see that the percentage of rational nodes in the system with the agent-based incentive mechanism is nearly lo%, while the percentage of rational nodes in the system with no incentive mechanism is nearly 70%. for the curve of the incentive P2P system -+Stands for the curve of the non-incentike P ~ svstem P 7
2006
abroad to study. As more scholars studying incentive mechanisms for P2P systems, we believe that there will be more developments. To solve the problems in P2P systems, such as Free-riding and Tragedy of Commons, in this article, we propose an agent-based incentive mechanism for P2P systems. By analyzing the experimental results, we conclude that it solve the fair sharing problem of information and resources in the P2P systems to some extent. With the aptitude of the agents, we can solve effectively the problems existing in the P2P systems. However, there are also problems we have not encountered yet. In future, we will conduct further research of the incentive mechanism, such as the research of reputation issues. Acknowledgements This work is sponsored by the National Natural Science Foundation of China (60573141, 70271050), National 863 High Technology Research Program of China (2005AA775050), the Natural Science Foundation of Jiangsu Province (BK2005 146), High Technology Research Programme of Jiangsu Provinc (BG2005038, BG2006001), High Technology Research Programme of Nanjing (2006RZ105) and Key Laboratory of Information Technology Processing of Jiangsu Province (kjsO5001, kjsO6).
+Stands
x
References I.
Buragohain C, Agrawal I),Sun S. A game theoretic framework for incentives in P2P systems peer-to-peer Computing. Proceedings of 3rd International Conference on Peer-to-Peer Computing, Sep 1-3, LinKoping, Sweden. Los Alamitos, CA, USA: IEEE Computer Society Press, 2003: 48-56
2.
Singh M. Peering at peer-to-peer computing. IEEE Internet
3.
Computing, 5 (6): 4-5 Matei R, Lamnitchi A, Foster I. Mapping the Gnutella network.
4.
Adar E, Huberman B A. Free riding on Gnutella. Technical
IEEE Internet Computing, 2002,6 (1): 50-57 I
0
20
40
60
8
I
80
100
Report, SSL-00-63. Palo Alto, CA, USA: Internet Ecologies Area Xerox Palo Alto Research Center. 2002
Simulation timesitn Fig. 2 The change of the percentage of rational nodes
5.
From the above simulation experimental results, we can see that with the agent-based mechanism, the P2P system can stimulate the peers to share information and resources with others. In this way, the information and resources sharing problems of P2P networks can be solved.
5 Conduslons and future work With the pervasive deployment of computers, P2P is increasingly receiving attention in research, product development, and investment circles. Research on incentive mechanism of P2P systems is very significant. It is important for the development of the P2P systems. And it has abstracted the scholars of home and
Hardin G. The Tragedy of the Commons. Science, 1968, 162: (3859) 1243-1248
6.
Feldmany M, Laiz K. Quantifying disincentives in peer-to-peer networks.
Proceedings of Workshop on Economics of
Peer-to-peer Systems, Jun, 2003, Berkeley, CA, USA. Berling, Germany: Springer-Verlag, 2003: 117-122 7.
8.
Su Jan-zong, Li Bing-zhi. Research of incentive mchanism in P2P file-sharing framework. Journal of Chongqing University of Posts and Telecommunications, 2006, 18(1): 123-125 (in Chinese) Xu Li , Zheng Bao-yu. Application and development of mobile agent in AN. The Journal of China Universities of Posts and Telecommunications, 2004, 11 (1): 73 -78
No. 4
ZHANG Jim-feng, et al.: An agent-based incentive mechanism for P2P systems
Shneidman J, Parkes D C. Rationality and self-interest in peer to peer networks. Proceedings of Second International Workshop on Peer-to-peer Systems (IF’TPS), Feb 20-21, Berkeley, CA, USA. Berling, Gemany: Springer-Verlag, 2003:47-52 10. Ma R T B, Lee S C M, Lui J C S, et al. An incentive mechanism for P2P network. Proceedings 24th International Conference on Distributed Computing Systems (ICDCS’04), Vol 24, Mar 24-26, Tokyo, Japan. Piscataway, NJ, USA: IEEE Computer Society, 2004:516-523 11. Feldman M, Lai K, Stoica I, et al. Robust incentive techniques for peer-to-peer networks. Proceedings of the 5th ACM Conference on Electronic Commerce: Vol 5, May 17-20, 2004, New York, USA. New York, NY, USA: ACM, 2004: 102-1 1 1 12. Feigenbaum J, Shenker S. Distributed algorithmic mechanism design: recent results and future directions. Proceedings of the 6th International Workshop on Discrete Algorithms and Methods for Mobile Computing and Communications (DIAL-M 2002), Sep 28,2002,Atlanta, CA, USA. New York, NY,USA: ACM,2002: 1-13 13. Buse D P, Sun P, Wu Q H, et al. Agent-based substation automation. IEEE Power and Energy Magazine, 2003, l(2):
9.
50-55
85
14. Wooldridge M. An introduction to multi-agent systems. England Chichester, U K John Wiley & Sons, 2002:31-45 15. Huang Hai-ping, Wang Ru-chuan, Xu Xiao-long. Research on applications of cryptography for multi-mobile agent system. Journal of China universities of Posts and Telecommunications, 2004,11 (4):29-32 Biographies: ZHANG Jian-feng, female, Postgraduate student in Nanjing University of Posts and Telecommunications. Major in computer software, intelligent agent technology, p2P technology and so on.
WANG Ru-chuan, male, professor of College of Computer of Nanjing University of Posts and Telecommunications. Tutor of doctorial graduate studento, Major in computer software, computer network, E-commence, network security and mobile agents.