The Evolution of Cloud through SJF-ML Hybrid Scheduling

Abstract

Purpose: The author proposes sixteen Shortest Job First - Machine Learning (SJF-ML) hybrid algorithms, combining the cloud's SJF scheduling algorithm with four ML algorithm categories, with cloud evolution through ML intelligence as the primary objective. The four categories include: SJF-CA, SJF-ELA, SJF-PM, and SJF-RA. The developed SJF-ML algorithms by the author perform pattern recognition of the tasks that are to be computed, to improve decision-making during task computations in the cloud. These sixteen SJF-ML algorithms include: SJF-ADAB, SJF-BAY, SJF-DT, SJF-KNN, SJF-LAS, SJF-LDA, SJF-LGB, SJF-LN, SJF-MLP, SJF-NAV, SJF-PLY, SJF-RDG, SJF-RF, SJF-RBST, SJF-SVM, and SJF-XGB. Performance Metrics: Cost, Time, Energy, and LB are utilized to compare the developed algorithms with baseline SJF, along with comparing them within their respective SJF-ML categories. Dataset: The real-time Google Big Data Task (BDT) dataset, comprising tasks ranging from one hundred to one thousand across nineteen files, was computed using the SJF-ML and SJF algorithms. Experiment: Open-source CloudSim simulator with VM counts of 20, 40, 60, 80, and 100 were utilized to compute the BDTs, outputting results across the considered metrics. Results: The algorithms SJF-XGB and SJF-LN provided the best results, with SJF-DT, SJF-LAS, and SJF-LDA providing poor results. Findings: Hybridization of the cloud's scheduling algorithms with ML provides improved intelligence and performance, resulting in the evolution of the cloud.