Improving Cloud Task Scheduling in Cloud Sim Plus using Demand-Aware VM Selection

Abstract

Deep reinforcement learning and metaheuristics are becoming popular ways to study cloud task scheduling to make things like makespan and completion time better; however, these methods often require a lot of computing power and can make it harder to understand and repeat the results. This paper suggests a simple Demand-Aware VM Selection policy for CloudSim Plus (simulated tasks) that improves the basic scheduling process by better matching tasks with virtual machines based on task needs, while leaving the rest of the simulation the same. The method was tested against the default time-shared baseline with three workload sizes (200, 500, and 1000 cloudlets) and six random seeds (42-47). The experimental results demonstrate that the proposed method drastically shortened the overall time of task completion: it raised the average time by 35.65% for 200 cloudlets (A computational task/work unit submitted for execution is CloudSim abstraction) and 28.86% for 500 cloudlets. In terms of average completion time, the newly planned lowered the average by 26.68% at 200 cloudlets and 28.53% at 1000 cloudlets; on the other hand, the 500-cloudlet scenario showed very high variability and even a slightly negative average improvement (-1.42%), indicating that completion-time averaging was sensitive under that workload regime even though there was a strong makespan gain. The results show that demand-aware binding is a clear, repeatable, and easy-to-add improvement for scheduling in CloudSim Plus, which can serve as a better starting point or as part of more advanced optimization and learning systems.