ARA: A Robust Audit to Prevent FreeRiding in P2P Networks

1
ARA A Robust Audit to Prevent Free-Riding in P2P
Networks

• MyungJoo Ham and Gul Agha
• Open System Laboratory
• University of Illinois at Urbana-Champaign

2
Outline

• Motivation
• Related Work
• Assumptions
• Key Idea
• Design
• Analysis
• Future Work

3
Motivation Free Riding
29099 (87) Contributed NOTHING

• More Contribution ? More Throughput?
• Some systems have anti free-riding mechanisms.
• However

7349 Shared, but NO responses
21750 Shared NOTHING!
11585 shared at least one file
11585
4236 responded at least once
From Free Riding on Gnutella by E. Adar and B.
Huberman
4
Motivation Cheating

• Throttle down outgoing traffic.
• Crack variables in the memory.
• Copy/Modify local files.
• Malicious Peers!

5
Outline

• Motivation
• Related Work
• Assumptions
• Key Idea
• Design
• Analysis
• Future Work

6
Related Work

• Counted by Each Peers
• Depends on clients authenticity
• eMule/eDonkey/Pruna
• Prone to bandwidth throttling,
• local file copy/modification
• Counted by Servers
• Need servers
• Usually detection is not accurate
• Usually depends on clients authenticity.

7
Related Work

• Reputation Sharing
• Popularity and Authenticity of shared resources
• Quality of resource
• Human intervention
• Not concerned in quantity
• MicroPayments
• Need servers
• Mental transaction costs

8
Outline

• Motivation
• Related Work
• Assumptions
• Key Idea
• Design
• Analysis
• Future Work

9
Assumptions

• Pure P2P
• Selfish peers
• Public Key Infrastructure
• ID on each peer (volatile)
• Fairness
• Being served according to its contribution

10
Assumptions Types of Cheating
I have 1,000.00. But, peer 3 lies that I have
0.00.
I have 1,000.00 ! Ask peer 2!

• Exaggerated Credit
• Conspiracy
• Collaborated
• Blame Transfer
• Blame innocent peers
• Omitting Interested Peers
• Hide witnesses

Hey! I have 1,000.00 !
Yes, peer 1 should have 1,000.00 !
11
Outline

• Motivation
• Related Work
• Assumptions
• Key Idea
• Design
• Analysis
• Future Work

12
Kea Idea

• Credit
• Contribution - Consumption
• Ability to consume others resource
• Volatility
• Recent information only
• Forget old information

13
Key Idea

• Interested Peers of peer i
• Peers had transactions with i recently
• Affected if i cheats
• Interested in i
• Audit the integrity of i
• I.Ps keep information of i
• Utilize PKI
• Others have is information signed by i

14
Key Idea

• Reward
• Credit over Lower Limit deserves services
• Priority for more credit value
• Punishment
• Banned
• Blacklist disseminated to others (I.P.s)
• Clues signed by the cheater needed

15
Outline

• Motivation
• Related Work
• Assumptions
• Key Idea
• Design
• Analysis
• Future Work

16
Design Credit

• Ci(t) Credit value of peer i at period t

• Fast Start-Up
• Grows faster until kC LLe

Volatility From t-m To t-1 During recent m
Contribution - Consumption

• LLe effective Lower Limit
• Ci(t) lt LLe ? service denied

17
Design Credit Behavior
Download more when you have more credit
Upload lt Download slightly
Credit
Upload Download
Upload gt Download
Lower Limit
Time
18
Design Interested Peers of peer i

• Have Ci(t)
• Have the list of interested peers of i
• Have the transaction records of i
• Response to queries about i
• Check the integrity of i

19
Design Integrity Check by peer j for peer
ICredit Audit
I have 500
Im interested in i I will check Ci
Credit of i ?
I have 500
Happens pC
Im interested in i
Peer i said its 500
20
Design Integrity Check by peer j for peer
iTransaction Record Audit
Happens pT
T.R. with j, k1, k2, , kn
Im interested in i I will check T.R. with k1, k2
T.R. with i ?
TR
TR
Chosen pTS
T.R. with i ?
Im interested in i
Im interested in i
21
Design Integrity Check by peer j for peer
iInterested Peers List Audit
Happens pI
Im interested in I Ill Check I.P. list
I.P.list of i
Check returned I.P. list Choose k1, k2 as samples
yes/no
Chosen pIS
j in your I.P. list of i ?
I.P.list of i
yes/no
I.P.list of I (partial)
I.P. of i ?
j in your I.P. list of i ?
P2P System
I.P.list of i
Im interested in i
Im interested in i
22
Outline

• Motivation
• Related Work
• Assumptions
• Key Idea
• Design
• Analysis
• Future Work

23
Analysis To Show Performance

• Detection Rate
• detected
• Overhead
• Bytes per period (10min)
• Detection Rate vs Overhead
• With varying parameters
• pT, pTS, pI, pIS Execution/sampling probability

24
Analysis Exaggerated CreditTransaction Record
Audit
1000

• Sampled Point
• pT 0.10, pTS 0.05
• Overhead
• 726 bytes/period
• Detection Rate
• gt98.5 for Exaggerated Credit
• gt97.5 for Conspiracy

0.10
Varying execution probability constant pT
0.08
Varying sampling probability constant pTS
0.10
0.06
0.09
Overhead
0.04
0.08
0.07
0.06
0.02
0.05
0.04
0.03
0.02
0.01
Detection Rate
100
25
Analysis Omitting Interested PeersInterested
Peers Audit

0.10
0.06
140

• Sampled Point
• pI 0.10, pIS 0.05
• Overhead
• 125 bytes/period
• Detection Rate
• gt98.5 for Omitting Interested Peers

0.05
Varying execution probability constant pI
0.05
0.04
Varying sampling probability constant pIS
0.04
Overhead
0.03
0.03
0.02
0.02
0.01
Detection Rate
100
26
Analysis Conclusion

• Detection Rate
• Exaggerated Credit gt98.5
• Conspiracy gt97.5
• Omitted Interested Peers gt98.5
• Overhead
• 154 2 bytes/sec
• 92600 851 bytes/period

27
Outline

• Motivation
• Related Work
• Assumptions
• Key Idea
• Design
• Analysis
• Future Work

28
Future Work

• Without PKI
• For smaller and less powerful nodes
• Distributed PKI may introduce longer latency
• Based on voting
• Reduce overhead
• Caching and Multicast
• Huge redundancy!
• Quality of Resource

29
Future Work

• Performance Measurement
• Fairness
• Comparison!
• Dynamic environment
• Simulation with Game Theory
• To predict users/peers reactions
• To see the popularity of systems with different
  strategies

30
Typographic Errors

• Page 1 at the top
• University of Illinois as Urbana-Champaign
• as ? at
• Page 7 at the top-left
• Transaction Credit Audit
• Credit ? Record