Limiting Duplicate Identities in Distributed Systems

1
Limiting Duplicate Identities in Distributed
Systems

• Elliot Jaffe, Dahlia Malkhi, Elan Pavlov
• The Hebrew University of Jerusalem
• Presented at
• Future Directions in Distributed Computing
• Survivability Obstacles and Solutions
• Bertinoro, Italy, June 23-25 2004

2
Things to come

1. What are Self-Replication attacks
2. Why Trust systems are not sufficient
3. How to leverage routing to allocate IDs
4. Defenses against self-replication attacks

3
Sample System

• Peers join and leave the system at will
• System performs work by distributing operations
  between peers
• Peer(s) should not be able to hijack the system
• Results should be safe under a threshold of
  colluding nodes

4
Attack Mode

• What if
• The number of bad peers is unbounded
• Bad peers could collude at no cost
• How would this effect the system?

5
How-to collude for free

• Self-Replication attacks
• A single node A sequentially joins the system
  multiple times, creating virtual identities
  A1..An
• All these identities share data
• Unbounded number of identities over time
• Instant zero cost collusion
• The Sybil Attack Douceur 2002

6
What is Survivability?

• Availability of System and its Data
• Distributed Control and Confidentiality

7
Availability

• Typical System
• P2P File Sharing
• Attacks
• Directed DOS attacks on key components
• Preventing access to data
• Delivering corrupted data

8
Availability Defenses

• Trust Relationships
• Damiano, di Vimercati, Paraboschi 2002
  Reputation based resources
• Kamar, Schlosser, Garcia-Molina 2003 EigenTrust
  Reputation Management
• Singh, Liu 2003 TrustMe Management of Trust
  Relationships

9
Trust Example
C

• B,C,D interact successfully with A
• E wants to talk to A
• E asks around about A
• B,C,D report positive experience
• E talks directly to A

B
A?
D
A?
A?
A
A - OK
A - OK
A - OK
E
10
Trust Based Defenses

• Trust is based on past performance
• Anonymity is important
• Hide ownership of files
• Hide targets from DOS attacks
• Multiple identities
• influence trust values
• hide source of corrupt files

11
Distributed Control and Confidentiality

• Distributed Key Escrow, Secure Storage
• Eternity, OceanStore, Farsite
• Attacks
• Directed DOS attacks on key components
• Preventing access to data
• Delivering corrupted data
• Exposure of confidential data

12
Secret Storing Defenses

• Trust based approaches
• provide integrity but not confidentiality
• there is no way to assign reputation to lurkers
• We need to use a pro-active approach to identify
  self-replicated nodes

13
Background Cryptographic Challenges

• First Proposed in 1975 by Merkle
• Bounded by available resources
• Since a node can physically perform at most C
  challenges, a self-replicated node will fail one
  or more challenges

14
Basic Model

• Every node has limited resources
• May be static or renewable
• Node can solve C challenges in time T
• Full Mesh connectivity

15
Simple Approach
C 2
1 A challenges B
C
2 B challenges A
B
D
Limits identities
2
Limits In/Out Degree
1
Repeat Challenge every T time
E
A
Assumes interactions gtgt T
Does not require global revocation
F
H
G
16
Probabilistic Approach
C 2
Every T, Choose a random node to challenge
C
B
D
Limits identities
Repeat Challenge every T time
E
A
Requires global revocation
A
F
H
G
17
Routable Identities

• Address is a routing ticket
• Routing paths define and limit the available ID
  space
• Attempted IDs outside of the scope are not
  routable

18
Routable Addresses Examples

• Internet
• Class C 192.168.5.
• 128 Routable addresses
• Ad Hoc Networks
• Paths through network
• Defines how packets are routed

19
Sisyphus Approach
C
Setup 1. Choose ID as Routable Address
B
A?
D
A?
2. Hash to find Vouchers
A?

• Runtime
• A wants to talk to E
• E asks As vouchers about A
• Vouchers challenge A
• A solves challenges
• Vouchers maintain certificate
• E talks directly to A

A
A - OK
A - OK
A - OK
E
ID(A) XYZZX
H1(ID(A)) B H2(ID(A)) C H3(ID(A)) D
20
Sisyphus

• No limit to in/out degree of interaction
• Does not require a revocation service
• Self-replication nodes fail challenges

21
Sisyphus Issues

• How do you identify actual vouching nodes based
  on hash value?
• Values do not correspond to actual nodes!
• Requires a weakly trusted component
• ID is defined by Routable Address
• Realistic

22
Summary

• Self-Replication attacks
• Trust systems are not sufficient
• Leverage routing to defining IDs
• Defense against self-replication attacks

23
Thank you