Abstract

The study focused on the growing need to manage the increasing workload pressure that often occurred in modern data centers during peak hours. The rapid growth of digital services had increased the computational demand, and this situation had created significant stress on resource utilization, energy consumption, and task scheduling. This background highlighted the need for a data-science-driven mechanism that handled workload patterns in an adaptive and efficient way. The problem centered on the fact that conventional scheduling techniques rarely adapted to irregular spikes, and many of these techniques have handled clustered loads poorly, which caused delays and underutilized resources. The method introduced an improved Density-Peak Adaptive Clustering (DPAC) algorithm that used recent advances in unsupervised learning and that analyzed dynamic workload traces collected from heterogeneous servers. The algorithm calculated local densities, identified core points, and formed adaptive clusters that represented different workload intensities. The model then mapped these clusters to appropriate resource pools, and it balanced the load across the data center. The framework also included a predictive module which has used historical patterns to anticipate the next peak interval. Experimental tests were carried out on a real workload dataset that included web services, database transactions, and analytics jobs. The proposed DPAC framework improves performance and efficiency of data centers during peak workloads. Experimental results indicate that the method reduces response time to 165 ms at 100% CPU utilization, while achieving CPU and memory utilization of 92% and 95%, respectively. Energy consumption decreases to 95 kWh, and the load balancing index reaches 0.75, demonstrating a significant improvement over k-means, reinforcement-learning-based allocation, and Gaussian mixture model approaches. These findings indicate that the framework has provided an adaptive, predictive, and energy-aware solution for optimized workload allocation in heterogeneous data centers.

Authors

P. Santhi, Kulangara Silpa Prabhu, T. Akhila Arjunan, A.J. Jiji
IES College of Engineering, India

Keywords

Data Science, Adaptive Clustering, Workload Allocation, Data Center Optimization, Peak-Hour Management