ADAPTIVE EDGE-BASED BANDWIDTH OPTIMIZATION FOR EFFICIENT DATA FLOW AND CONGESTION CONTROL

Abstract

This study proposes an adaptive edge-based bandwidth optimization system to enhance network efficiency, accelerate data transfer, and mitigate congestion. The system leverages a hybrid reinforcement learning approach integrated with an Adaptive Federated Bandwidth Allocation (AFBA) mechanism to dynamically manage resources. The Deep Q-Network (DQN) framework optimizes bandwidth allocation by learning from real-time network fluctuations, while the Attention-based Graph Neural Network (AGNN) enhances predictive accuracy by analyzing traffic patterns across distributed nodes. The implementation incorporates Rust for high-performance concurrency, Apache Cassandra for scalable distributed storage, and Envoy proxy for efficient inter-node communication. Extensive simulations conducted under dynamic network loads validate the effectiveness of the proposed system. Performance metrics such as Bandwidth Utilization Efficiency (BUE), jitter, end-to-end delay, and packet delivery ratio confirm superior adaptability and responsiveness compared to existing centralized and decentralized models. The results emphasize the advantages of edge intelligence in achieving enhanced scalability, reduced congestion, and optimized resource allocation, making the proposed approach ideal for high-demand and latency-sensitive applications.