Secure network coding with a cost criterion

Abstract

Network coding is a viable alterative to traditional routing methods that are being used in current data networks as it offers many advantages, including higher throughput rates and lower network costs. Although research has been conducted with regards to the cost and security issues of network coding, a joint investigation of these two parameters has not been done yet, thus providing the motivation for this thesis. For this thesis, we consider the situation where a set of messages is to be multicasted across the network, of which a known subset is of interest to a wiretapping adversary. The problem that we attempt to solve is to find a network coding scheme that has both a low network cost and a low probability of the wiretapper being able to retrieve all the messages of interest. We make use of random linear codes in anticipation for decentralized implementation of the scheme, and focus on the problem of finding the multicast subgraph. As an exact algorithmic solution is difficult, we propose two heuristic solutions, and compare their performances to traditional routing through a simulation study on Rocketfuel networks. Our results suggest that network coding can be more effective than routing for this low cost and secure data multicast problem, especially when the links are not easily tapped.