A Collaborative NDN Caching with Multi-Agent Deep Reinforcement Learning and Blockchain Incentives

Abstract

As data-driven applications and user demands grow, managing content delivery in Named Data Networking (NDN) has become more challenging. Traditional caching methods struggle to scale in dynamic and decentralized environments where content popularity changes and collaboration among routers is needed. This paper introduces a decentralized collaborative caching framework for NDN, combining Multi-Agent Deep Reinforcement Learning (MADRL) and blockchain technology. MADRL enables routers to autonomously adjust caching strategies based on local states and interactions with neighboring routers, improving cache hit rates and reducing retrieval costs. Blockchain technology ensures fair and transparent rewards through a cryptocurrency-based token system, incentivizing collaboration and minimizing free-riding risks. The framework also integrates Delegated Proof of Stake (DPoS) for efficient, secure validation of caching actions. Simulations demonstrate that the approach significantly enhances caching efficiency, reduces latency, and improves scalability, addressing the challenges of dynamic decentralized environments.