A load-aware replica selection strategy with multi-armed bandits and adaptive redundancy in ICN
Wancai Zhang, Jianping Song, Rui Han · International Journal of Information and Communication Technology · 2026
With the rapid growth of data-intensive applications, achieving low-latency and reliable content retrieval in complex networks has become a major challenge.Information-centric networking (ICN) leverages content naming and pervasive in-network caching to enable retrieval from multiple replicas, making replica selection crucial for performance.However, selection is complicated by replica capacity limits, bursty workloads, and dynamic path variations.To address these issues, we propose a replica selection strategy that integrates the multi-armed bandit (MAB) framework with dynamic redundancy control.By modelling selection as an MAB problem, the strategy incorporates path variability, service heterogeneity, and blocking risk into decision-making, enabling adaptive exploration and exploitation.An additional load-aware redundancy mechanism adjusts redundancy levels to curb exploration overhead and suppress tail latency.Simulations on a real-world topology show that the method significantly reduces latency and improves robustness.Compared with nearest-replica routing, it reduces average latency by 32.09% and P99 tail latency by 45.76%.