Moving Target Defense Against Cache Pollution Attacks in MEC-based Metaverse

To realize metaverse services, achieving low latency is essential. Caching content on nodes closer to the users in mobile edge computing (MEC) networks serves as an effective strategy to meet this demand. However, the cache strategy usually leads to cache pollution attack (CPA). CPA causes the edge nodes to be overwhelmed with unpopular content, which results in users not being able to access content from the edge nodes, thus making the latency increase again and affecting the metaverse service. Although existing defense strategies work, they are mostly reactive and suffer from untimely defense. Considering this problem, we divide user-node connections into actual request connection (ARC) and logical cache connection (LCC) and propose a proactive defense policy that selects LCC by analyzing group ARC in each time slot. By shuffling LCC, the policy selects the appropriate node for the content to be cached to achieve attack content aggregation. We model the connection change process as a Markov decision process (MDP), which is then solved by deep reinforcement learning (DRL). Through numerical simulation experiments, the proposed algorithm outperforms the reactive defense strategy.