An Energy‐Efficient Load Balancing Framework Using Edge Servers in Software‐Defined Networks
Lokesh Pawar, Gaurav Bathla, Rohit Bajaj · International Journal of Communication Systems · 2025
ABSTRACT Software‐defined networks empower the networking world by cutting short the requirements of network hardware equipment. Hardware load management equipment is error‐prone and costly. SDN uses a smart controller that is readily available for the network and releases the burden of the traditional network. Clustering is utilized to form several types of suitable clusters of network nodes. A double clustering formulation is proposed in the article to reach an optimized solution. Before using the smart controller, the spectral clustering algorithm is used to improve the clustering of the nodes. Load management is achieved with the help of optimized clustering and load balancing using edge servers based on throughput, priority‐based, and queue‐based request resolution. The authors have proposed an algorithm for load balancing in TCP and UDP types of traffic and a combination of both types of traffic. The bandwidth of the network plays a vital role in the performance of the network; taking bandwidth as a major variable, various observations regarding performance are framed in the proposed work, which effectively represents good performance when compared with similar strategies. A new policy for load balancing, a queuing model, is framed for the tasks generated in the network. Low compute intensive (LComi) task or low task (LT) and super low task (SLT) are directed to edge servers, and high compute intensive (HComi) or super high task (SHT) and high task (HT) are directed to cloud servers for managing the latency of the network. Overall, a 12.22% improvement is found in the throughput of the network, considering time and bandwidth as a major sources of input to the cluster classification, priority assignment, and load balancing algorithm. A total cumulative 1% and 4% improvement is found in energy consumption using tree and single topology metrics.