Dynamics of Learning Algorithms for the On-Demand Secure Byzantine Routing Protocol

We investigate the performance of of several protocol enhancements to the On-Demand Secure Byzantine Routing (ODSBR) [3] protocol in the presence of various Byzantine Attack models. These enhancements include a) Nodal Weighting (in addition to Link Weighting) in the reputation database, b) a network layer retransmission capability and c) modifications to the packet flow rates. These enhancements are meant to improve the learning rate of the protocol in the presence of various Byzantine Attack models. The attack models investigated include previously investigated models [4] and a new and effective attack model, termed the MAC-Level Attack. We investigate the dynamics of the ODSBR protocol and its enhancements through analytic models and extensive simulation studies. We find that the protocol enhancements improve the learning times of the ODSBR protocol. The Nodal Weighting enhancement specifically helps in the presence of the various colluding Byzantine Attack models investigated. However, the time to develop a relatively complete reputation database in the MANET context is found to be relatively large.