Abstract
Despite recent research efforts, wireless ad hoc networking technology remains especially prone to security attacks. In this work our contribution focuses on determining the optimal trade-off between traffic dispersion and TCP performance to reduce the chances of successful eavesdropping, while maintaining acceptable levels of throughput. For our experiments we propose a multipath-enhanced version of DSR, and we compare Tahoe, Reno and Sack TCP variants. Results show that multipath traffic dispersion impact on TCP throughput is bounded to a maximum of 25-35%, and that there is only a minimal dependence on the number of routes used, the number of consecutive packets sent on each route, the route selection algorithm or the TCP variant used.