Nature Inspired Energy Efficient Task Scheduling Algorithm based on Multi Criteria Decision Making method in Cloud Computing

Abstract

Cloud Computing is a revolution in the IT industry which provides on demand services to the users based on the SLA made between the Cloud user and Provider. Task scheduling is one of the facet in the cloud computing. It is a prodigious challenge in Cloud Computing. The main objective of Task scheduling in Cloud Computing is to map the incoming tasks onto the virtual resources which were running in physical hosts running in datacenters while minimizing the makespan. In the existing works, many of the task scheduling algorithms were developed but focused only on the metrics like makespan, Execution cost. The goal of the thesis is to develop a multi objective scheduling algorithm which aims to address the metrics makespan, memory utilization, Bandwidth utilization, migration time, migration cost, Power cost and energy consumption by using the nature inspired algorithms. In this work we have proposed an algorithm which schedules the incoming tasks on to the virtual machines by calculating both the Task and VM priorities while minimizing the makespan, migration time, Energy consumption, power cost and maximizing the resource utilization in terms of bandwidth and memory utilization. This algorithm is modeled by using the nature inspired algorithms i.e, PSO, CS and crow search algorithms. newlineWe have carried out this research by modeling with the nature inspired algorithms namely PSO, CS and Crow search algorithms. Totally we have carried out this research in two stages. In the stage-1, initially we have modeled the algorithm by hybridizing PSO and Cuckoo search by considering the constraints Task and VM priorities and it is compared with the existing ACO, GA, PSO and CS algorithms and the proposed approach is outperformed interms of the above mentioned metrics. In the stage-2, we have modeled the algorithm with the crow search algorithm and compared with the existing ACO,PSO and CS algorithms and the proposed approach is outperformed the existing algorithms in terms of the above mentioned metrics. newlineFor the first stage we have considered the workload from the parallel workload archives from HPC2N and NASA logs and the random workload generated from the planet lab in the cloudsim. For the second stage we have considered the random workload generated from the planetlab in the cloudsim. The simulation is carried out in the cloudsim and after completion of the two stages we have identified that our proposed approaches outperformed the existing algorithms i.e, ACO, PSO, CS interms of the above mentioned specified metrics. newline