1

A Social Network Integrated Reputation System for Cooperative P2P File Sharing.

協同 P2P 文件分享的社交網路綜合聲望系統Author(s)

Kang Chen Dept. of Electr. & Comput. Eng., Clemson Univ., Clemson, SC, USA Haiying Shen ; Sapra, K. ; Guoxin Liu

Published in:

Computer Communications and Networks (ICCCN), 2013 22nd International Conference on

Date of Conference:

July 30 2013-Aug. 2 2013

報告者：

Ma1g0222 童耀民 2014/10/01

2

Outline-------------------------------------------大綱

3

Outline

Introduction.

The design of socialtrust. A. Social Networks. B. Lightweight Reliable Server Selection. C. Reputation Evaluation.

Performance evaluation. A. Efficiency of Reputation Systems. B. Effectiveness of Reputation Systems.

Conclusions.

4

Introduction-------------------------------------------介紹

5

Introduction

Due to the open nature of the peer-to-peer (P2P) environment, P2P file sharing systems are prone to have selfish and misbehaving nodes.

Selfish nodes are not cooperative in providing files, but still would like other nodes to comply to their requests [ 1 ], [ 2 ].

Misbehaving nodes can distribute tampered files, corrupted files or files with malicious code into the system, which could be further spread by unsuspecting users.

cooperative

6

Introduction

Therefore, incentives are needed to encourage cooperation in P2P networks.

Reputation system, as a cooperation incentive method, has been widely studied in recent years [ 3 ] [ 4 ] [ 5 ] [ 6 ] [ 7 ].

7

Introduction

However, a clever node can sustain in the system by maintaining its reputation just above the threshold and take this advantage for uncooperative behaviors.

Further, frequent reputation querying can easily overload the reputation center, leading to degraded service quality in P2P systems.

8

Introduction

Recently, emerging P2P file sharing systems have been proposed to incorporate online social networks (OSNs) to enhance service cooperation [ 8 ] [ 9 ] [ 10 ] [ 11 ] or malicious node detection [ 12 ] by leveraging the social property of “friendship fosters cooperation” [ 13 ].

Naturally, such an idea can alleviate the necessity of reputation querying and reduce the load on reputation centers.

9

Introduction

In this paper, they propose a credit based reputation system, namely SocialTrust, that synergistically integrates the traditional reputation systems and social networks to overcome their individual shortcomings with three main components:

(1) Social networks (friend network and partner network) (2) Lightweight reliable server selection (3) Reputation evaluation for cooperative file serving and honest rating

10

The design of socialtrust.-------------------------------------------Socialtrust 的設計。

11

A. Social Networks. (社交網路 )

B. Lightweight Reliable Server Selection.(輕型可靠性服務器的選擇 )

C. Reputation Evaluation.(聲望評估 )

12

A. Social Networks

In a general OSN, a user's friends include offline friends with certain social connections (eg, friends, classmates, colleagues, etc.) in real life and online friends.

Similarly, the social relationships of a node in SocialTrust include both offline friends and trustable online partners.

13

A. Social Networks

1) Friendship Maintenance.

Since the friendship is user centric, each user maintains its own friend-list.

When a node, say Ni, wants to add another node, say Nj , into its friend-list, it sends a friend invitation to Nj. If Nj accepts the invitation, they become friends of each other.

If a user deletes a friend, they remove each other from their friend-lists.

14

A. Social Networks

2) Partnership Maintenance..

The interaction frequency between the two nodes is larger than a threshold, denoted by .

Each node's reputation value is larger than the partnership threshold, denoted by .

15

A. Social Networks

3) Further Discussion.

A friend or a partner may also be uncooperative.

This strategy, in turn, provides cooperation incentives to friends and partners since fewer friends and partners would lead to fewer opportunities to provide service and earn reputation.

A node would be regarded as a selfish node if its reputation drops below a pre-defined threshold.

Then, its services will be rejected by other nodes in the system.

16

B. Lightweight Reliable Server Selection

Since the friendship and partnership represent trust, they exploit this property to alleviate the reputation querying cost.

Each node also maintains local ranks for friends/partners based on previous service records to handle the case when when several friends or partners appear in the available server list.

17

B. Lightweight Reliable Server Selection

Since the friendship and partnership represent trust, they exploit this property to alleviate the reputation querying cost.

18

B. Lightweight Reliable Server Selection

They then deduce the percentage of reputation queries that can be avoided in SocialTrust. 

Then, the probability that none of the available servers is a friend or partner

是由節點 產生的伺服器請求。 伺服器請求的平均數。 是 friends and partners node 數量。

19

B. Lightweight Reliable Server Selection

should be larger.

As shown in Equation (1) , the more friends/partners a node has, the more reputation queries (ie, cost) it can avoid.

As stated in Section III-A , the friendship is usually stable while the partnerships are built dynamically.

20

C. Reputation Evaluation.

The reputation system updates node reputations based on received reputation feedbacks, which determines the accuracy and the effectiveness of the incentive system in encouraging cooperation and discouraging non-cooperation.

1) Cooperative Server and Cooperative Client

2) Cooperative Server and Non-cooperative Client

3) Non-cooperative Server and Cooperative Client

4) Non-cooperative Server and Non-cooperative Client

21

C. Reputation Evaluation.

Therefore, we propose a metric called Trust (denoted by) ) that integrates both reputation and social degree:

is the reputation of node , is the maximal reputation value allowed in the system.

is the social degree of , is the maximal number of friends and partners a node can have in the system.

is an adjusting factor.

22

C. Reputation Evaluation.

1) Cooperative Server and Cooperative Client

23

C. Reputation Evaluation.

2) Cooperative Server and Non-cooperative Client

24

C. Reputation Evaluation.

3) Non-cooperative Server and Cooperative Client

25

C. Reputation Evaluation.

4) Non-cooperative Server and Non-cooperative Client

26

D. Performance evaluation.

a. 成本比較b. 檢測正確的非合作節點c. 檢測虛假的非合作節點d.(d) plots the number of non-cooperative services received by all nodes.

27

D. Performance evaluation.

3: Effectiveness of reputation systems

28

D. Performance evaluation.

4: Accuracy in reputation evaluation of cooperative nodes

29

D. Performance evaluation.

5: Accuracy in reputation evaluation of non-cooperative nodes

30

E. Conclusions.

本篇論文，提出了 P2P 網路的 SocialTrust 系統，其中為了節省聲望查詢成本，提高聲望評估的準確性，整合社交網路特性的社交網路聲望系統。

朋友 / 社交夥伴通常是互相信任的所以可以減少聲望查詢的成本。