RANKING AND REPUTATION BASED RESOURCE ALLOCATION IN P2P SYSTEM

Ranking and Reputation Based Resource Allocation in P2P System

Omaia Mohammad Awad Al Omari

, Nayyar Ahmed Khan

and Rund Mahafdah

Lecturer, Department of Computer Science and IT, Shaqra University, Shaqra, Saudi Arabia. Email: rundmahafdah@su.edu.sa

Article Received: 11 October 2017 Article Accepted: 30 November 2017 Article Published: 22 December 2017

1. INTRODUCTION

In peer-to-peer networks, nodes can be both resource providers and resource consumers at the same time. In this

sense, the services offered by a peer-to-peer network rely on resource sharing among peers. This work focuses on

how peers share their access link capacity between upload and download rates. In our model peers are rational

agents, and choose their strategy in order to maximize their own utility. Most of the users of p2p systems free ride,

i.e., consume resources. This is also known as a social phenomenon reported as “tragedy of the commons” that

most of the users are reluctant to cooperate and only a small number of them are willing to share their resources

without contributing any.

Many methods have been used in the literature to motivate users to cooperate, like reputation based methods [3],

ranking based method [1], pricing-based schemes, game theoretical methods [5], and secure key issuing scheme. In

this paper, we give the appropriate incentives for cooperation in systems of peers who use a single capacity limited

access link both for uploading and downloading content. In this we use the concept of reputation and ranking based

optimal resource allocation in p2p network [3]. First, we will discuss about reputation policy, in this each peer act

as client and server simultaneously. Reputation is a value which lies between 0-1. It is provided to a peer in the

network according to its performance which is based on uploading and downloading.

When a peer provides its bandwidth to any other peer for uploading then its reputation value increases. Thus

contributive, peers are rewarded by receiving preferential treatment, while misbehaving peers are punished by not

being served. Here we propose an overall allocation framework according to which each peer independently

decides how to allocate his available resources. On one hand, reputation-based allocation policies are followed

based on which each peer determines the quality of service that he will offer to each one of his requesters according

to their reputations and demands. On the other hand, rational strategies are used, based on which peers, being aware

A B S T R A C T

Most structured Peer-to-Peer networks assume that all nodes in the network are altruists. However, as the networks grow larger, they may include many selfish nodes, which leech resources from the network without contributing any in return. This is a serious problem in existing P2P systems. Thus, as a solution to this problem we propose two new schemes. One of them is Distributed Reputation-based System, according to which peers earn reputation in direct accordance to their contributions to the system. In this way, each peer has to manage a tradeoff between the bandwidth capacity he will dedicate for uploading (in order to increase his reputation or ranking) and the bandwidth capacity he will dedicate for his downloads, other scheme is the Ranking based in which we define a utility function which capture the best wish for the source peer to serve competing peers, who request services from the source peer.

of the reputation system’s existence, dynamically adapt the capacity that they dedicate for uploading and

downloading in order to improve their utility. Our proposed allocation scheme is implemented in a distributed

manner at each peer independently of the others and the only information that is passed from one peer to another is

the requested amount of resources. The allocation scheme is combined with a reputation-based server selection

policy to help peers select among the most reputed/contributive servers and avoid misbehaving (non-contributive)

ones. Now let us discuss our second strategy i.e. based on Ranking. Here we consider a resource allocation

mechanism, which allows source peers to provide competing peers differentiated services [1]. Source peers provide

services and allocate their resources to serve competing peers. These competing peers consume services and use the

resources allocated by source peers.

A ranking mechanism is incorporated in the resource allocations. The rankings are derived from peer’s behavioral

histories of providing services and consuming services. Therefore, a competing peer with a good history of serving

other peers is provided a service with better quality, because the source peer will allocate more resources to serve

the peer. This service differentiation provides incentives for peers to contribute their resources to P2P systems. In

the context of resource allocation problems, fairness is an important criterion for mechanisms that allocate shared

resources. The remainder of this paper is organized as follows: Section 2 summarizes and compares related work,

section 3 describes the Resource allocation problem Section 4 describes the system model. In Section 5, we propose

a capacity allocation algorithm for rational peers along with three possible peers’ request strategies, in section 6 we

evaluate the problem by the two strategies (Ranking and Reputation. Finally, Section 7 concludes the paper.

2. RELATED WORK

As we had discussed reputation and raking based methods before, now we will understand some more methods that

can be implemented [2]. One of them is the “Incentive Scheme for Optimizing Network Performance in Structured P2P Networks”. In this to remove the problem of free ride: a virtual currency scheme is proposed. In such a scheme,

each node is associated with a certain amount of virtual “money”. A node could earn money for providing service

and has to pay for the services that it consumes. Therefore, any node that refuses to provide service will end up with

no money and then cannot consume service any more. The purposes of existing virtual currency schemes are to

encourage all nodes to take the role of servers. However, the servers in such schemes are only concerned about how

to maximize their own revenues, regardless of whether their consumers are satisfied with the receiving services.

This conflicts with the original intention of P2P network— that is, the network should harness all available

resources in the network efficiently to maximally satisfy the needs of users. Therefore, the network performance

under these schemes may be suboptimal from the perspective of users.

To achieve this goal, we apply a market model in which nodes participate and we leverage the concept of price to

regulate the behaviors of nodes. In the market model, a server has the freedom to specify the service prices and the

bandwidth allocation for its clients. A client selects its favorite server and pays a price to the server for every unit of

Other related work in this area is on “Game Theoretic Framework for Incentives in P2P Systems” which is used to

study the interaction of strategic and rational peers, and propose a differential service-based incentive scheme to

improve the system’s performance [5].

There is another scheme which gives a secure key issuing scheme for p2p network (SKIP). Current IBC (Identity

based cryptography)-based solutions could not address the problem of secure private key issuing. SKIP proposed a

peer registration solution using Shamir’s (k, n) secret sharing scheme, and a secure key issuing scheme, which

adopts key generate center (KGC) and key privacy authorities (KPAs) to issue private keys to peers securely in

order to enable the IBC systems to be more acceptable and applicable in real-world P2P networks. The theoretical

analysis and experimental results show that SKIP performs effectively and efficiently, and is able to support large

scale networks.

3. RESOURCE ALLOCATION PROBLEM

The resource allocation proposed utilizes publicly observable and verifiable information to achieve optimal

resource allocation [1]. The ranking incorporated in the resource allocation is derived from peer’s behavioral history, which is a peer’s history of providing services and consuming services. When a source peer allocates

resources to serve other peers, we assume the source peer makes decision independently. We also assume that one

peer in our model is the source peer who provides the services. Other peers are competing peers, who consume the

services provided by the source peer. Let set M be the set of competing peers, and

M = |M| is the number of these peers. The competing peers compete to share the source peer’s resources.

The ranking pi> 0 pi of peer i is given by

pi = f (ci , si )

Where f (ci , si ) is the ranking function, which is increasing in ci and decreasing in si . The ci and si are peer i’s

histories of consuming services and providing services respectively.

Therefore, a good friend is rated with a small number and total stranger is rated with a large number. Note that a

suspicious malicious peer could also be rated with a large number and hence the ranking has the potential to be used

for alleviating malicious attacks, such as flooding attacks. A friend is a peer who has an observed history to serve

the source peer or other peers. Therefore, the higher pi is, the less peer i has contributed to the P2P system.

The source peer should allocate xi unit of resources to peer i and Ri>=xi>=ri>=0. Assume that the resource capacity

of the source peer is C > 0. The minimum resources r

r = (r1,L,rN ) has to satisfy the source peer’s capacity constraint, i.e., We call Ri the target resources.

Here the parameter α > 1 represents the sensitivity to the deviation from target resources. As depicted in Fig, the

utility is more sensitive to deviations when α is large and vice versa. This utility function reflects that the source

peer wishes to serve the peer with resource Ri since Ui (xi )→∞ when xi → Ri . When xi deviates from Ri , the

utility decreases sharply because the peer’s quality of service deteriorates.

The utility function simply says that the source peer wishes to serve the competing peer with the target resources.

4. PROPOSED MODEL

Our focus here is how to dynamically allocate in a distributed way the resources (bandwidth in this case) of the p2p

community among their members in a way to provide fairness and efficiency in the system by guaranteeing that

peers will be able to receive resources in proportion to their contributions, and that all available resources will be

fully exploited. In p2p network peers act as both client and server. We consider that each peer i is connected to the

Backbone Network through an access link with capacity Ci (measured in b/s), reflecting his available bandwidth for

uploading and downloading content [3]. We consider that our system progresses in periods of a fixed number of

time units. Each peer decides about the capacity that he will dedicate in a given period p for uploading to others (we

will refer to it as upload capacity), uCpi, and the capacity that he will use for his own needs (we will refer to it as

download capacity), dCip . It is obvious that .uCpi + dCip = Ci for any p and i.

At the beginning of a period p, each peer decides where to send his gi requests (randomly or according to our

reputation based server selection policy) and the amount of bandwidth he will request from each one of his servers

under Basic, Greedy, or Adaptive strategy. During his first download period in the system, a peer sets dCip and

uCip =0. This corresponds to either newcomer, i.e., peers who joined the system for the first time in order to

download something or old peers who were idle for some time (not downloading) and decided to begin

downloading again. We will refer to both newcomers and such old peers as beginners.

Each peer allocates his current upload capacity u CIP over the peers’ requests that were directed to him. If the

peer is a beginner, he does not serve any request (u Ci p = 0). Allocation is performed through our proposed

reputation-based allocation policies. At the end of the period p, each peer calculates the reputation of his servers

the next period p+1, based on the capacity allocation strategy described in Algorithm. Our model considers duplex

connections, as it determines how a peer allocates his physical capacity among his downlink and uplink streams,

which take place simultaneously [3].

5. RATIONAL STRATEGIES

Capacity Allocation Strategy: Peers who try to maximize their utility with the least possible contributions, by

progressively increasing their upload capacity till they obtain the quality of service they desire [3]. Therefore, if the

bandwidth a given peer I receives during a period p (Bi p) smaller than the capacity that he dedicates for

downloading in the given period, it means that he needs to increase his cooperation level in order to improve his

reputation and be able to receive more resources from the network. So, in this case, peer I will increase his upload

capacity by a constant bi as soon as it does not exceed his total capacity and decrease the download capacity by the

same constant bi. On the other hand, if the peer receives as much bandwidth as his download capacity can afford, he

will try to maximize his utility by further increasing his download capacity by the constant bi for the next p þ 1

period, in cost of decreasing his upload capacity by the same constant bi. From our simulations, we saw that a good

trade-off between convergence rate and performance is succeeded in the system when each peer I uses a bi in the

order of Ci=10. Aforementioned algorithm is presented below.

Basic Strategy (BS)

In this, each peer I demands from each one of his servers in a given period p bandwidth equal to dCIP/ gi]. Since

peers split their demands uniformly between servers and are unaware of the other peers’ load, they may face the

situation under which one of their servers cannot serve them and another one could give them more than what they

actually asked. Since servers do not allocate more resources than the reported demands of the peers, peers may

receive much less bandwidth than what they could in a given period. In order to confront this problematic case, we

propose the greedy strategy.

Greedy Strategy (GS)

If the upload capacity of a peer i is not fully exploited (because the demands of the competing peers are less than his

available upload capacity), he uses the residual capacity for downloading [3] Each peer i requests from each one of

his server’s bandwidth equal to his total current download capacity. This, he increases the possibility of receiving

more bandwidth from an available and capable peer. When more than 1 server provides the peer its requested

Adaptive Strategy (AS)

In this, each peer I requests from each one of his server’s bandwidth equal to his total current download capacity.

However, server knows its requesters capacity. For example, if server sends with a rate of 9 Mb/sec but the

download capacity of peer is only 7 Mb/sec, then server receive acknowledgement of sent packet with a rate of

7Mb/sec [3]. Therefore, he gives rest of the bandwidth to the other peer.

6. EVALUATION

Suppose that there are four competing peers that request services from the source peer,

Case: 1

We assume that there are four competing peers and the resource requirements are R = (1, 2, 3, 4) and r = (0.25, 0.5,

0.75, 1). Suppose that all the peers have the same ranking. A linear increase when the source peer contributes more

resources (i.e., increases the resource capacity) [1]. Therefore, every competing peer perceives a better service

when the source peer increases the resource capacity.

Case: 2

Suppose that resource requirements of four competing peers are R = (1, 4, 1, 4) and r = (0.25, 1, 0.25, 1), the first

two peers have the same rankings and the other two peers’ rankings are also the same but different from the first

two peers [1]. The rankings of peer 1 and peer 2 are better than the rankings of peer 3 and peer 4. That is, the first

two peers are friends and the other two peers are strangers of the P2P system. The resources allocated to a friend as

shown in Fig. is higher than that to a stranger, although they have the same resource requirements. For strangers, we

see that they suffer more if their resource requirements are higher. Therefore, there are strong incentives for a peer

Case: 3

Now we assume that four competing peers require the same resource Ri = 1 and ri = 0.25, for I =1, 2, 3, 4. However,

their rankings are different, that is, p = (1, 2, 3, 4). As we see in Fig, when the resource capacity is just enough to

satisfy the minimum requirements of these four peers, every peer is allocated the minimum resources [1]. As more

resource capacity becomes available, friends are given higher priority in distributing the extra available resources.

However, this cannot go beyond the target resources. When the resource capacity approaches total amount of all

target resources, the differences between friends and strangers diminish. It consists with that the source peer should

satisfy all competing peers with the target resources when the source peer is not congested.

Evaluation 2

In this, we exhibit the behavior of the system under the case of new peers periodically entering and leaving the

network [1]. Here we consider a network of 100 peers where peers are categorized according to their total physical

capacity; there are five equal-sized categories of peers with 8, 7, 6, 5, and 4 Mb/s capacities. However, we consider

that 50 percent of each category peers are permanent in the system (stable), while the remaining 50 percent of each

category (capacity) peers stay in the network only for 100 periods and are replaced by new identity peers of the

same capacity with those who left every 100 periods. We use this case in order to keep the same analogy of capacity

peers in the system and better investigate the performance of the system compared to the static one where no

arrivals or departures occur. In Fig, we can see the performance of the stable and newcomers of 8 and 4 Mb/s

The conclusions from the performance of the other capacity peers and when the GS/RA-SS or AS/RA-SS strategies

are used, remain the same. What we see from Fig. is that the performance of the stable peers is not much influenced

by the presence of the newcomers, and on the other hand, newcomers’ behavior is very close to the one of the stable

peers of the same capacity, despite their short life in the network. This conclusion, which is also supported by many

different scenarios that we run, shows that newcomers adapt quickly to the network and do not really affect the

performance of the long-lived peers in the system.

7. CONCLUSION

The limitation of Bit Torrent is as follows, it assumes a clear distinction between the upload and the download

capacity of the peers. The download capacity is much bigger than the upload capacity, which is considered to be the

bottleneck in the connections [3]. If BT peers had to share a limited single-link capacity both for uploading and

downloading data, they should decide how to allocate it between their uplink and downlink connections. Current

implementations of BT do not support a capacity adaptation algorithm to help peers dynamically alter their

download and upload capacities in order to improve their performance. The problem with that is that if all peers

acted rationally, they would rather use their whole capacity solely for downloading and the system would collapse,

to overcome these problems this gave rise to the two main strategies that we discussed in our paper based on

Reputation and Ranking based resource allocation system. In this paper we motivated the peers to contribute their

resources in the network where capacity limited access link of each peer is being shared among the download and

uploads. We considered rational peers who seek to maximize their utility with the least possible contributions and

showed that under the presence of our proposed reputation system they are inclined to cooperation. The optimal

resource allocation in the “Ranking based system” achieves max-min fairness, which maximizes the minimum

social welfare obtained. Rankings are incorporated in the resource allocation for the source peer to provide

differentiated services [1]. A peer improves his/her ranking by contributing resources to the P2P systems and

deteriorates his/her ranking by consuming services. A peer with bad ranking is allocated fewer resources than peers

with better rankings so that the peer perceives a worse quality of service

REFERENCES

[1] Ranking-based Optimal Resource Allocation in Peer-to-Peer Networks Yonghe Yan, Adel El-Atawy, Ehab

[2] An Incentive Scheme for Optimizing Network Performance in Structured P2P Networks Guowei Huang,

Zonghai Li, Zhi Chen, Gongyi Wu Department of Computer Science Nankai University, China

[3] Reputation-Based Resource Allocation in P2P Systems of Rational Users Anna Satsiou, Student Member,

IEEE, and Leandros Tassiulas, Fellow, IEEElinger.

[4] Building Heterogeneous Peer-to-Peer Networks:Protocol and Analysis Kin-Wah Kwong, Student Member,

IEEE, and Danny H. K. Tsang, Senior Member, IEEEJ.

[5] A Game Theoretic Framework for Incentives in P2P Systems � Chiranjeeb Buragohain Divyakant Agrawal

Subhash Suri Computer Science Department University of California, Santa Barbara, CA 93106

[6] Bandwidth Trading in Unstructured P2P Content Distribution Networks Kolja Eger and Ulrich Killat

Department of Communication Networks Hamburg University of Technology (TUHH)

[7] D. Hughes, G. Coulson and j. Walkerdine, “Free Riding On Gnutella Revisted The Bell Tolls,” IEEE

Distiributed Systems Online, vol. 6, no. 6 June 2005.