DAG-Hash Load Balancing Technique with Green Servers for Data Centers Networks
Abstract: This paper presents a dynamic, energy-aware data distribution and load balancing framework tailored for Software-Defined Networking (SDN)-enabled data centers. The proposed system leverages a novel DAG-Hash mechanism. It is a weighted consistent hashing integrated with DAG to allow connecting nodes in a hash ring as a DAG. The DAG connection of nodes is used to replace a hash ring and allow simplicity and rapid node connection without needed to rebuild the hash ring when a node’s weights are updated. The node weights are updated based on real-time performance metrics, including CPU utilization, memory consumption, throughput, and energy efficiency. This adaptive approach ensures equitable load distribution while minimizing reconfiguration overhead during network changes. To enhance energy efficiency, the framework incorporates a threshold-triggered power management protocol that transitions underutilized servers into sleep mode once data replication requirements are met and reactivates them on demand. A Directed Acyclic Graph (DAG) is constructed periodically to visualize node interactions and weight propagation, enabling real-time monitoring and decision support. The system is prototyped using Mininet and the Ryu controller, with realistic traffic simulated via the GWA-Bitbrains dataset. By simulating the traditional methods, including standard consistent hashing, weighted least connection and other algorithms, to approve the validation for the proposed system, many experimental scenarios had been concerned. The results show that the proposed approach significantly outperforms them in terms of throughput, latency, CPU and memory load distribution, and energy efficiency. Experimental results display that DAG-Hash improves throughput by 22.6%, reduces response time by 31.1%, and lowers Power Usage Effectiveness (PUE) by 5.5% in comparison to existing schedulers which include WLC, traditional constant hashing and EC-Scheduler. Moreover, DAG-Hash achieves higher Data Center Energy Productivity (DCEP) by using as many as 40 units under high traffic. The proposed approach additionally performs more efficiently on Scafida topology, which suggests a decrease in PUE and a higher DCEP in comparison to Fat-Tree under similar load situations. These results confirm DAG-Hash as a scalable, electricity-green, and topology-adaptive answer for wise traffic control in software program-defined statistics centers.