1
SybilGuard Defending Against Sybil Attacks via
Social Networks

• Authors Haifeng Yu, Phillip B. Gibbons, and
  Suman Nath
• (several slides based on authors)

2
The Problem

• Redundancy lets distributed systems compensate
  for faulty nodes
• Ex Store data on multiple nodes
• The Sybil Attack undermines redundancy
• Need a central authority to determine which nodes
  are honest

3
SybilGuards Central Authority

• Main Idea Use a social network as the central
  authority
• A node trusts its neighbours
• Each node learns about the network from its
  neighbours

4
Sybil Nodes and Attack Edges
Attack Edges
honest nodes

• - Edges to honest nodes are human established
• - Attack edges are difficult for Sybil nodes to
  create

5
Attack Edges Are Rare

• SybilGuard hinges on having relatively few attack
  edges
• To subvert system an attacker must compromise
  many honest nodes

6
SybilGuards Model

• A social network exists containing honest nodes
  and Sybil nodes
• Honest nodes provide a service to or receive a
  service from nodes that they accept
• Ideally, only honest nodes are accepted

7
SybilGuards Guarantees

• With high probability an honest node
• Accepts most honest nodes
• Is accepted by most honest nodes
• Accepts at most a bounded number of Sybil nodes
• (Can partition accepted nodes into sets, of which
  a bounded number contain Sybil nodes)

8
Segue Random Routes

• Every node picks a random routing from input to
  output edges
• A directed edge is in exactly one route of
  unbounded length
• A directed edge is in at most w routes of length w

e
9
Clever Use of Random Routes

• Each node finds all the length w random routes
  that start at it
• Honest node V accepts node S if most of Vs
  random routes intersect a random route of S
• Why does this work?

10
Random Route Intersection Honest Nodes

• WHP
• verifiers route stays within honest region
• routes from two honest nodes intersect

Verifier
Suspect
sybil nodes
honest nodes
11
Random Route Intersection Sybil Nodes

• Each attack edge gives one intersection
• Intersection points are SybilGuards equivalence
  sets

Verifier
Suspect
same intersection
sybil nodes
honest nodes
12
Nodes Accepted per Intersection
Verifier

• Verifier accepts at most w nodes per intersection

for a given intersection
13
Bounds on Accepted Sybil Nodes

• For routes of length w in a network with g attack
  edges, WHP,
• Accepted nodes can be partitioned into sets of
  which at most g contain Sybil nodes
• Honest nodes accept at most wg Sybil nodes

14
Applications of SybilGuard

• Can SybilGuard be applied to any current
  distributed systems?
• Does it allow any new systems to be created?

15
Restrictions Imposed On Applications

• There must be a social network
• Nodes must create and maintain their friendships
• How many social networks will we need?
• One for each application, or
• A single network used by many applications

16
Privacy Implications

• Information about friends spreads along routes
• Verification involves nodes sharing all their
  routes
• Bloom filters help here
• Nodes are not anonymous