1. A generic framework towards trust building in self-organized, peer, networks
Giannis F. Marias, Vassileios Tsetsos,
Odysseas Sekkas and Panagiotis Georgiadis
Dept. of Informatics and Telecommunications, University of Athens
SecPerU 2005
July
Santorini, Greece

2. Presentation Structure
Introduction
ATF Architecture
Simulations, Future Work and Conclusions
SecPerU 2005, Santorini, Greece

3. Self-organized networks
MANETs
Ad hoc collaborations
No infrastructure available
Many threats from selfish, malicious or hacker nodes
Advanced needs for QoS and security
SecPerU 2005, Santorini, Greece

4. Trust Management
A new paradigm for security and QoS solutions in open systems
Involves (among others):
recommendations exchange
reputation building/fading
Does not require central authorities
Many different trust management schemes have been proposed
SecPerU 2005, Santorini, Greece

5. Motivation
Trust management schemes seem suitable for ad hoc collaborations
Those proposed for MANETs are too specialized
Those proposed for middleware services are too complex to apply to MANETs
Belief networks, probabilistic methods
A lightweight flexible framework is needed for assessing the trustworthiness of nodes
 ATF (Ad hoc Trust Framework)
SecPerU 2005, Santorini, Greece

6. Presentation Structure
Introduction
ATF Architecture
Simulations, Future Work and Conclusions
SecPerU 2005, Santorini, Greece

7. Network and Application Stack
Overall Architecture
ATF is a trust evaluation and advertisement framework
NOT a cooperation enforcement technique
NOT a security framework
TS1
TS2
Trust Builder
Trust Matrix
Reputation Manager
Trust Policy
Network and Application Stack
RFTS
SecPerU 2005, Santorini, Greece

8. Trust Sensors Every node provides functions to other nodes
Packet forwarding, routing, naming services, …
Trust Sensors evaluate the quality of these functions in a node’s neighborhood – i.e., capture the direct evidence
Observation of neighbors’ behavior
Comparison to reference/ideal behavior
Quantification of the difference to Success/Failure
SecPerU 2005, Santorini, Greece

9. Reputation Manager On-demand recommendations exchange
The nearest and most trustworthy recommenders are selected based on the TrustSpan Algorithm
Recommendations are requested only when there are insufficient direct evidences
SecPerU 2005, Santorini, Greece

10. TrustSpan algorithm
Packet forwarding for one- and two-hop neighbors is monitored through AODV headers and MAC overhearing
At most N trusted recommenders are selected
They are detected with a special TS, called RFTS (Rec. Function Trust Sensor)
The algorithm is invoked periodically and only for target nodes with less than M direct interactions
SecPerU 2005, Santorini, Greece

11. Trust Builder Main components:
Direct evidence (DE)
Recommendations (REC)
History of interactions
Subjective factor (SUB)
The values for all open parameters are defined in the Trust Policy of each node
Trust Values are assigned per (node, function) in a Trust Matrix
SecPerU 2005, Santorini, Greece

12. Trust Computation (I)
Time is discrete and counts separately for each (node, function) according to the direct evidence
w and H are defined in Trust Policy so as to decrease the trust fluctuations without losing sensitivity
SecPerU 2005, Santorini, Greece

13. Trust Computation (II)
α, b, SUB: defined in Trust Policy
SecPerU 2005, Santorini, Greece

14. SUB SUB is a time function in the range [0,2]
It allows for the introduction of subjective criteria in trust assessment
SUB=0  distrust always
SUB=1  use the default ATF trust scheme
SUB=2  be enthusiastic
Ideally used for modeling more complex time-variant behaviors and trust strategies
Example strategy: do not trust the function X of any node until there are W successful interactions
SecPerU 2005, Santorini, Greece

15. Presentation Structure
Introduction
ATF Architecture
Simulations, Future Work and Conclusions
SecPerU 2005, Santorini, Greece

16. Simulations (I)
We have already performed some simulations using J-Sim and AODV as the routing protocol
We deal with the packet forwarding function
The first results are quite satisfactory regarding overheads of TrustSpan and identification time for selfish nodes
Simulation results will be published soon
SecPerU 2005, Santorini, Greece

17. Simulations (II) f = packet forwarding
#nodes = 50
max speed = 1.2 m/s
packet rate = 4 pkts/s
area = 300x300 m2
Averaged identification times for 10 selfish nodes
SecPerU 2005, Santorini, Greece

18. Future Work Test ATF with more high-level functions/protocols
Assess its robustness and accuracy through extensive simulations
Investigate the effects of individual trust strategies in the trust establishment process (implemented through the SUB component)
SecPerU 2005, Santorini, Greece

19. Conclusions
ATF is a generic framework in a sense that is function-agnostic
It is lightweight as it entails only simple mathematical computations
It involves a subjective component as dictated by trust-related research in other disciplines
e.g., psychology, cognitive science
The large number of Trust Policy parameters enable flexibility in trust building
SecPerU 2005, Santorini, Greece

20. Thank you for your attention!!!
The end
Thank you for your attention!!!
Questions???
SecPerU 2005, Santorini, Greece