P2P network traffic control mechanism based on global evaluation values

P2P network traffic control mechanism based on global evaluation values

The Journal of China Universities of Posts and Telecommunications June 2009, 16(3): 66–70 www.sciencedirect.com/science/journal/10058885 www.buptjour...

227KB Sizes 0 Downloads 111 Views

The Journal of China Universities of Posts and Telecommunications June 2009, 16(3): 66–70 www.sciencedirect.com/science/journal/10058885

www.buptjournal.cn/xben

P2P network traffic control mechanism based on global evaluation values XU Xiao-long1, WANG Ru-chuan1, 2 ( ) 1. School of Computer, Nanjing University of Posts and Telecommunications, Nanjing 210003, China 2. State Key Laboratory for Novel Software Technology, Nanjing University, Nanjing 210093, China

Abstract

Peer-to-peer (P2P) computing technology has been widely used on the Internet to exchange data. However, it occupies much network bandwidth, and thus greatly influences traditional business on the Internet. Besides, problems about free-riders and ‘tragedy of the commons’ in the P2P environment estrange from it P2P users who constantly contribute to the network with quality resources. This article proposes a new P2P network traffic control mechanism based on global evaluation values. It aims to help individual users to avoid peak traffic time as much as possible, ease network congestion and protect traditional business on the Internet, as well as differentiating priority grades of peers according to their contributions and stimulating them to share their valuable resources actively. This article first analyzes the current state of network traffic, and then elaborates on P2P network traffic control policies and proposes the peer’s priority level differentiation mechanism based on global evaluation values. Finally, after the testing results and analysis of the proposed P2P network traffic control mechanism are discussed, conclusions are drawn. Keywords Peer-to-peer computing (P2P), traffic control, global evaluation values

1

Introduction 

In P2P computing environment, every individual is equal to each other in status, which means that they all can be service providers and consumers at the same time. This mode of computing is totally different from the traditional client/server (C/S) or browser/server (B/S) computing. P2P technology can be appropriately applied in large scale information sharing systems, instant message software and cooperating work platforms, etc [1–2]. The most popular applications of P2P are in the field of file sharing, such as BitTorrent, eDonkey, Gnutella, eMule which grow to be the biggest consumers of Internet resources [3–4]. The P2P file sharing system is typically characterized by online users constantly downloading large files from the Internet with P2P tools .These users exhaust shared resources without contributing any valuable information, which pulls the P2P system back to C/S or B/S mode and makes network routers and switches always work to full capacity [5–6]. Therefore, the Web, E-mail, and other traditional Internet Received date: 14-05-2008 Corresponding author: WANG Ru-chuan, E-mail: [email protected] DOI: 10.1016/S1005-8885(08)60229-0

services are affected seriously, and subsequently, peers who contribute considerable useful resources cannot be fairly rewarded. On this account, a novel P2P network traffic control mechanism is proposed in this article. To ensure the interests of high-quality peers and restrict P2P traffic to a reasonable state, this mechanism is designed to classify peers according to their owners’ behavior, and then control their actions based on their priority levels. Its most important element is global evaluation value.

2

Network traffic

The main services provided on the Internet now include the Web, E-mail, instant message (IM), file transfer protocol (FTP), video-on-demand (VoD), P2P file sharing services, etc. Statistical data show that P2P services occupy about 90% of Internet bandwidth, whereas among these online clients, only 25% are P2P users, about 65% are using web services and only getting 7% bandwidth, and about 50% are using IM services with only 1% bandwidth.(Source: www.internet2.edu July’04 and European Tier IISP Feb’04) The backbone network traffic of the Internet during

Issue 3

XU Xiao-long, et al. / P2P network traffic control mechanism based on global evaluation values

different periods of time fluctuates in a day, as shown in Fig. 1. However, large files (such as movie files) exchanged between peers make the network traffic of the Internet even worse. (Source: www.safenext.com)

Fig. 1 Network traffic of the Internet at different periods of time in a day

In P2P environment, an individual will not always contribute its resources, although it demands resources from other peers. Statistical data show that 60% resources of the P2P environment come from 1% peers, and most of the other peers only gain resources without providing any resource, which causes the P2P systems to lose their P2P specialty with no difference from the traditional C/S mode. More serious problems in almost all P2P systems are free-riders, tragedy of the commons, non-cooperation, distributed denial of services (DDoS), and virus. These happen because of the lack of efficient trust and incentive mechanisms.

3 P2P network traffic control mechanism based on global evaluation values 3.1

P2P network traffic control policy

Quality of service (QoS) refers to the capability of a network to provide better service to selected network traffic by technologies. Its primary goal is to provide priority over dedicated bandwidth, controlled jitter and latency (required by some real-time and interactive applications), and improved loss characteristics (http://www.c114.net/keyword/QoS). Another important goal is to ensure that providing priority for one or more flows does not fail the other flows. It is significant for P2P network to ensure its QoS to work successfully. An efficient traffic control mechanism is supposed to be used in P2P network to guarantee different dataflow transmission, at different QoS levels. The P2P network traffic control policy is made according to the following rules: 1) Peers are supposed to get higher priority and be granted greater rights to get what they need for contributing valuable resources actively and stably. 2) If a peer only downloads files or uses others’ resources, but seldom provides any valuable service, it will be granted with lower priorities.

67

3) P2P network traffic must be limited when the public network is highly congested. Peers with lower priority are supposed to be especially restricted. The main idea is to differentiate levels of services by specialized time in a day. 4) P2P network traffic must be confined locally, which means information communicated among peers must be kept in the local area of the network as much as possible to alleviate the pressure of the backbone network. The policy enables peers to avoid the peak time of network traffic to alleviate the network jam and reserve communication bandwidth for traditional network services, and to differentiate quality of services based on peers’ priority level. 3.2 Peers’ priority level differentiation mechanism based on global evaluation values An individual’s priority level is determined by its own behavior. If it always contributes valuable resources actively and stably, it will get high priority based on the evaluation of other peers interacting with it. In a distributed environment, such as P2P environment, peers may still rate each other after each transaction, as in the eBay system. For example, each time peer i gets service from peer j, it may rate the transaction positive or negative. The reasons why peer i may rate it negative may be that the service provided by peer j is bad or failed. However, the transaction is also affected by the performance of the other elements of peer i itself. Inspired by the EigenRep [7] model of Stanford, a new peer priority level differentiation mechanism based on global evaluation values is proposed. Suppose that in a P2P network, there are a series of interactive activities between peer i and peer j (peer i obtains service from peer j) in a certain period of time 'x . Peer i is allowed to store the number of successful transactions it has had with peer j, si , j , and the number of failed transactions it has had with peer j, f i , j . A normalized local direct evaluation value of peer j in 'x provided by peer i , vi , j , is defined as follows: ­ max( si , j  P fi , j ,0) ; ¦ max( si , j  P fi , j ,0) z 0 ° j ° ¦ max( si , j  P fi , j ,0) (1) vi , j ® j °0; s f max( P ,0) 0  ¦j i, j i, j ° ¯ This ensures that all values will be between 0 and 1. And if ¦ max(si, j  P fi, j ,0) 0 , vi, j should be set to zero. P is a j

very important factor in this formula, which is larger than 1 and used as a critical level for failure. Therefore, the matrix of the local direct evaluation value in P2P network, V , is as follows:

68

The Journal of China Universities of Posts and Telecommunications § v1,1 ... v1, j ¨ ¨ # ¨ vi ,1 ... vi , j ¨ ¨ # ¨v © n ,1 ... vn , j

... v1, n · ¸ # ¸ ... vi , n ¸ V (2) ¸ # ¸ ... vn , n ¸¹ As discussed earlier, the result of transaction is determined by both peers participating in this transaction, which means that the performance or other elements of peer i who gets service from peer j should be considered, that is, the global evaluation value of peer i. After the interactions of the last period time 'x 1 in P2P network, the authors can obtain each individual’s objective evaluation value, the global evaluation vector gi , as follows:

gi , 'x1

§ g1, 'x1 · ¨ ¸ ¨ # ¸ ¨g ¸ ¨ i , 'x1 ¸ ¨ # ¸ ¨¨ ¸¸ © g n , 'x1 ¹

gi evolves from gi , 'x1 to gi , 'x as follows: § v1,1 ... vi ,1 ... vn ,1 · § g1, 'x1 · ¸ ¨ ¸¨ # ¸¨ # ¸ ¨ # gi , 'x V T gi , 'x1 ¨ v1, j ... vi , j ... vn , j ¸ ¨ gi , 'x1 ¸ (4) ¸ ¨ ¸¨ # # ¨ ¸ # ¨ ¸ ¸ ¨v ¸ ¨¨ g v v ... ... n i n n n 1, , , © ¹ © n , 'x1 ¸¹ Every individual is assigned an initial global evaluation value g i ,0 when it joins a P2P network for the first time:

gi ,0

­ p,0  p  1; Peeri  SPC ® ¯1; Peeri  SPC

When the global evaluation values and grades of the peers are obtained, the next important question is how to store them. The answers are two methods: one is to choose from a number of reliable super peers to manage and store them; the other is to use the mechanism of distributed hash table (DHT) technology to store them in different places of the whole P2P network. 3.3

(5)

Super peer cluster (SPC) is a set including peers that are initial leading nodes existing in the P2P network from the very beginning. To enable the algorithm easily applicable, the authors classify the peers into five grades based on their global evaluation values, as shown in Fig. 2: Service-worst, Serviceworse, Service-modest, Service-better, and Service-best.

Fig. 2 Five grades of peers classified based on global evaluation values

Traffic control mechanism

According to both the QoS policy and the peer priority level differentiation mechanism, and global evaluation values discussed earlier, the authors define the form of dataflow of the P2P network, which contains all the necessary information traffic control needed, as shown in Fig. 3.

(3)

Then, after the interactions of the current period time 'x ,

2009

Fig. 3

3.3.1 peer

Form of dataflow of P2P network

The relative locations of source peer and destination

A P2P network can be divided into several groups (peer group, PG) by the physical location of peers, and then it is easy to calculate the relative distance from the source peer identifier (SPID) to the destination peer identifier (DPID). If data exchange just happens between peers in the same PG, no load will be added to the network backbone, that is, the load can be ignored. 3.3.2

Data type

There are two types of information exchanged between peers: signaling information, such as data indexing, data searching, peer or file location and routing information, and data with actual value, such as text, pictures, audio, and video information. These two types differ greatly in amount. 3.3.3

Priority levels of source peers and destination peers

Peers can be assigned different grades based on peer priority level differentiation mechanism. Then the directions of dataflow can be easily classified into four kinds by their grades: from peers of high grade to peers of high grade, from peers of high grades to peers of low grades, from peers of low grades to peers of high grades, and from peers of low grades to peers of low grades. Obviously, dataflow among peers of high grade has the highest priority. Now, the logical structure and the operating procedure of P2P network traffic control system are described, as shown in

Issue 3

XU Xiao-long, et al. / P2P network traffic control mechanism based on global evaluation values

Fig. 4.

Fig. 4

The logical structure of P2P network traffic control system

1) First, the current traffic analysis and monitor component gets information about how busy the current network is. Subsequently, the policy selection component chooses the appropriate policy from the library of traffic control policy. 2) When a dataflow arrives, the traffic identifier component instantly distinguishes the current dataflow based on the information of the library of traffic mode. The concrete methods may be the data packet payload recognition method and the network dynamic characterizing method, which can be used to distinguish P2P, Web, E-mail, IM, etc. If the data is of P2P type, it will be sent to the P2P network traffic analysis component to discriminate; otherwise, it will be discarded. 3) The P2P network traffic analysis component abstracts information of the data packet, which is useful for traffic control, including SPID, DPID, data type, source peer group (SPG), destination peer group (DPG), etc, and then sends them to the traffic control component. 4) The traffic control component determines how to deal with the current data (to transmit, delay, or discard) according to the information sent from the P2P network traffic analysis component and the traffic control policy selected from the library. Obviously, traffic recognition is the prerequisite for traffic control. The methods of P2P network traffic flow recognition, can generally be divided into two categories [4,8–9]: one is based on the useful payload of the P2P data packet, such as deep packet identification (DPI) technology and packet signature technology; the other is characterizing P2P network traffic with only the knowledge of network dynamics, such as (IP, Port), block size, flow connection patterns, etc.

4 4.1

Experiments P2P network simulation platform

J-Sim (http://www.j-sim.org/) is a kind of Java-based, open-source, real-time process-driven network simulation platform. Besides supporting IP network, J-Sim is able to support many other wired or wireless networks, such as Ad-hoc networks, wireless sensitive networks, and is easily

69

applied to simulate different network layers, different network structures and different network components. In this article, the authors present how to utilize J-Sim and further extend it with the common object request broker architecture (COBRA)based distributed component technology, thus building a P2P network distributed simulation platform. This new platform can effectively save processing and consumption of resources, differentiate nodes based on their abilities, and support the Monte Carlo method, multiple protocols, dynamic network topology changes, etc. It adopts Nam and TCL script as the P2P network simulation scene editor, to simulate events of networks dynamically, particularly P2P network traffic changes in file-sharing scenario. 4.2

Analysis of the experiment results

The experiments are aimed at testing the performance of the P2P network traffic control mechanism proposed in this article at different times. The authors select two groups of peers: one group is composed of peers of higher priority level, which means that they provide better services than those providing modest services, and the other group is composed of peers of lower priority level. The experiments are divided into two parts. In the first part, the authors just simulate the regularity of the changes of P2P network traffic on the Internet without considering the other traffic of the main traditional services and get an average dataflow change of two peer groups in a basic time interval (24 hours) with their P2P network traffic control mechanism, as shown in Fig. 5.

Fig. 5 The average dataflow change of two peer groups in a basic time interval (24 hours) with the P2P network traffic control mechanism

From Fig. 5, it can be seen that the P2P network traffic control mechanism based on global evaluation values is able to differentiate services to ensure peers’ quality of service according to their priority level. Second, the authors set up the traffic control policy according to the congestion condition of the network (to restrict P2P network traffic at the peak moments and to release the restriction of P2P network traffic during idle moments) and priority levels of peers. The experimental results are shown in Fig. 6.

70

The Journal of China Universities of Posts and Telecommunications

2009

(2006AA01Z201, 2006AA01Z439, 2007AA01Z404, 2007AA01Z478), the High Technology Research Program of Jiangsu Province (BG2006001), the High Technology Research Program of Nanjing (2007RZ127), Foundation of the National Laboratory for Modern Communications (9140C1105040805), and the Science & Technology Innovation Fund for Higher Education Institutions of Jiangsu Province (CX08B-085Z, CX08B-086Z). Fig. 6

The result of the second part of the experiment

Fig. 6 shows that the P2P network traffic control mechanism can adjust P2P network traffic effectively based on network control policy and traffic distribution of a day.

5

Conclusions

P2P network traffic control mechanism has becoming a popular research topic. Almost all network operators need to, figure out how to effectively restrict P2P network traffic. The P2P network traffic control mechanism based on global evaluation values proposed in this article is verified effective by simulations in controlling P2P network traffic, and can reasonably ensure peers of high priority the use of network resources. However, normal operation of the mechanism also needs other mechanisms to cooperate, such as peer-to-peer identity authentication mechanisms. However, the traffic control policy should be studied further to enable P2P network traffic control to be more smooth. Acknowledgements This work was supported by the National Natural Science Foundation of China (60573141, 60773041), the Hi-Tech Research Program of China

References 1. Parameswaran M, Susarla A, Whinston A B. P2P networking: an information-sharing alternative. Computing Practices, 2001, 34(7): 3138 2. Loo A W. The future of peer-to-peer computing. Communications of the ACM, 2003, 46(9): 5661 3. Karagiannis T, Broido A, Brownlee N, et al. File-sharing in the internet: a characterization of P2P traffic in the backbone. Technical Report. Riverside, CA, USA: University of California, 2003 4. Sen S, Wang J. Analyzing peer-to-peer traffic across large networks. IEEE/ACM Transactions on Networking, 2004, 12(2): 219232 5. Adarand E, Huberman B. Free riding on Gnutella. Technical Report CSL-00-3. Palo Alto, CA, USA: Xerox (Palo Alto Research Center) PARC, 2000 6. Feldmany M, Laiz K, Chuang J, et al. Quantifying disincentives in peer-to-peer networks. Proceedings of the 1st Workshop on Economics of peer-to-peer Systems (P2PEcon’03), Jun 56, 2003, Berkeley, CA, USA. LNCS 2735. Berlin, Germany: Springer-Verlag, 2003: 117122 7. Kamvar S D, Schlosser M T, Garcia-Molina H. EigenRep: reputation management in P2P networks. Proceedings of the 12th World Wide Web Conference (WWW’ 03), May 2024, 2003, Budapest, Hungary. New York, NY, USA: ACM, 2003: 123134 8. Karagiannis T, Broido A, Faloutsos M, et al. Transport layer identification of P2P traffic. Proceedings of the 4th ACM SIGCOMM Conference on Internet Measurement, Oct 2527, 2004, Taormina, Italy. New York, NY, USA: ACM, 2004: 121134 9. Hamada T, Chujo K, Chujo T, et al. Peer-to-peer traffic in metro networks: analysis, modeling, and policies. Proceedings of the 2004 IEEE/IFIP Network Operations and Management Symposium (NOMS’04): Vol 1, Apr 1923, 2004, Seoul, South Korea. Piscataway, NJ, USA: IEEE, 2003: 425438

(Editor: ZHANG Ying)