Towards Traffic Engineering in Anonymous Communication Networks: A Meta-Routing Learning Approach

With the increasing demand for user privacy and communication security, Anonymous Communication Networks (ACNs), which allow individuals to conceal their identities during communications, have become widely used. However, the encryption mechanism that hides the transmission source and destination from third parties in ACNs often leads to inconsistencies between measured and actual traffic demands, thus significantly influencing the routing decision-making processes of traditional centralized Traffic Engineering (TE) schemes. In this paper, we focus on achieving TE in ACNs by proposing a novel Meta-Routing Learning (MRL) approach, which aims to train a routing agent capable of quickly adapting to new anonymous traffic patterns using only a few training iterations and measurements. Specifically, to eliminate the impact of inconsistencies on routing policies, we propose routing optimization for actual traffic demands by managing the traffic distribution on links with heavy communication overhead. To rapidly adapt to new anonymous traffic patterns in real network environments, MRL employs meta-learning to train the routing agent on a wide range of anonymous traffic patterns offline and periodically fine-tunes the trained agent online. Comparative experiments on real network topologies and traffic traces demonstrate the superior performance of MRL in achieving load balancing for TE in ACNs.