On-demand link padding in traffic anonymizing

We consider the problem of using artificially generated cover traffic to hide the network paths used by an application against an adversarial observer who can monitor network traffic and perform traffic analysis. We propose that on-demand link padding be used on top of existing traffic anonymizing infrastructures such as those described in the Onion Routing Project or in the ANON Project. In on-demand link padding, cover traffic is generated dynamically only when real traffic is present. Moreover, the bandwidth usage is economical because we only generate a small amount of cover traffic above real traffic. We describe two types of on-demand link padding schemes and evaluate their performance by simulation. We report our findings based on the simulation results and quantify potential trade-offs in configuring a couple of important parameters. Finally, we provide a strategy for mitigating the potential problem in face of lost or corrupted control packets, to which the traditional recovery solutions based on packet retransmission are inappropriate for performance considerations.