Modified Simple Linear Regression in Load Balancing Using CloudSim

Luis William C. Meing; Arnemie B. Gayyed; Dionisio R. Tandingan Jr.

Luis William C. Meing College of Information Technology and Computer Science, University of the Cordilleras, Philippines
Arnemie B. Gayyed College of Information Technology and Computer Science, University of the Cordilleras, Philippines
Dionisio R. Tandingan Jr. College of Information Technology and Computer Science, University of the Cordilleras, Philippines

Abstract

Purpose – This study addresses load balancing algorithms facing challenges in predicting resource usage for load distribution by exploring the incorporation of Pythagorean means (arithmetic, geometric, and harmonic) into regression models to enhance prediction accuracy for CPU, RAM, and bandwidth utilization in cloud environments.

Method – Focusing on the Round Robin Scheduling Algorithm, this paper introduces modified simple linear regression (SLR) models that integrate these means. Integrating these in the computation of a trendline function, computing then comparing the residuals on a sample dataset generated in CloudSim, is performed. A K-means model was used for comparison as it too was used in other literature.

Results – The findings reveal that incorporating the harmonic mean into SLR significantly reduces the mean squared error (MSE) in predicting CPU and bandwidth usage, offering a more nuanced approach to load balancing.

Conclusion – These results highlight the potential of harmonic mean-based SLR in refining resource prediction algorithms, suggesting an avenue for future research in developing more adaptable and efficient load-balancing strategies in cloud computing.

Recommendations – The study can be further validated in environments outside of CloudSim and further improved by adding the complexity of other measures of central tendency or linear regression methods.

Research Implications – The study encourages further exploration into cloud infrastructure optimization, the development of ML- enhanced load-balancing algorithms, and the use of other statistical means in ML. This suggests a broader impact, indicating areas for future research for strategies in load balancing and resource prediction in cloud computing environments.

Practical Implications – The improved load balancing and usage prediction capabilities could benefit cloud service providers and end-users and lead to more reliable, scalable, and cost-effective solutions.

Author Biographies

Luis William C. Meing, College of Information Technology and Computer Science, University of the Cordilleras, Philippines

Luis William C. Meing is a faculty member of the University of the Cordilleras – College of Information Technology and Computer Science. Having an undergraduate degree in Information Technology, they are expanding their horizons by pursuing a master's in Computer Science. This background has empowered them to go beyond the realm of ERPs to the algorithmic details that make them function.

Arnemie B. Gayyed, College of Information Technology and Computer Science, University of the Cordilleras, Philippines

Arnemie B. Gayyed is a faculty member of the University of the Cordilleras – College of Information Technology and Computer Science.

Dionisio R. Tandingan Jr., College of Information Technology and Computer Science, University of the Cordilleras, Philippines

Dionisio R. Tandingan Jr. is a faculty member of the University of the Cordilleras – College of Engineering and Architecture.