PongCache Poisoning In GUESS

1
Pong-Cache Poisoning In GUESS

• Neil Daswani
• Network Security Lab
• NTT DoCoMo USA Labs

Hector Garcia-Molina Computer Science Dept. Sta
nford University
2
P2P Protocols
3
P2P Security

• Research Challenge
• Autonomous, Distributed, Untrusted Nodes
• Malicious nodes indistinguishable from good
• Approach
• Load balancing / Provide fairness

4
Context Security Mechanisms
5
Containment
6
GUESS
A
C
CCS 04
B C
AB
7
GUESS Ping/Pong
A
C
pong C
B C
AB
8
Liveness Problem
9
Introduction Protocol
B
AC
A
C
Q
B C
AB
10
Malicious Nodes

• Propagate their IDs in pong-caches(poison
  caches)
  
• Denial-of-Service
• Inauthentic Results
• Further propagation
• Collusion
• Passive Active Propagation
• Techniques
• IDSA ID Smearing Algorithm
• DNP Dynamic Network Partitioning

11
ID Balancing
B
A D
A
C
AB
BE
D
E
C E
CD
12
ID Balancing
B
A D
A
C
AB
BM
D
M
C M
13
ID Smearing Algorithm

• At each node if id in a pong is already in
  pong-cache, remove it.
  
• Result
• Overload balances out id distribution
• DoS Attack prevents any particular malicious id
  from becoming widely replicated

14
IDSA Performance
How to deal with more malicious nodes?
Poisoning mitigated here.
15
Dynamic Partitioning

• n nodes
• m malicious(randomlydistributed)
• Cache Size
• Search onepartition at atime.

16
Dynamic Partitioning
Partition Number
j-bit Node Id (a)
?
j-1
p-1
0
Each node chooses a random ? and ?.
17
Dynamic Partitioning
Partition Number
j-bit Node Id (a)
?
j-1
p-1
0
Limits poisoning (on average) to
m2j-p
18
IDSA DNP
p 4
After Partitioning
Before Partitioning
19
QA

• Google Neil Daswani (links to publications on
  Research part of the page)
  
• Email daswani_at_cs.stanford.edudaswani_at_docomolabs
  -usa.com