| Cloud computing is a technology that provides a platform for sharing resources such as software,infrastructure,application,and other information,through which resources can be provided on demand to promote information technology and Industrial change.Among them,a data center(server infrastructure)hosts hundreds or thousands of servers consisting of software and hardware to respond to customer requests.Obviously,performing operations requires a large amount of energy.The energy consumption of all data centers around the world has been calculated to be 1.4% of the total electrical energy consumption and is increasing by 12% every year [1].Therefore,it is necessary to optimize energy consumption in data centers by understanding the energy flow and its distribution.There is an urgent need to develop various related efficient task scheduling algorithms to optimize execution time,cost,and energy consumption while maximizing resource utilization and system performance.Here,the development of task scheduling algorithms with excellent performance has always been a research hotspot in the field of cloud computing technology.We have noticed that most of the existing cloud computing task scheduling algorithms still have a lot of room for improvement in terms of intelligence and reasonable task allocation according to users’ QoS requirements to save computing resources.Moreover,with the increase of heterogeneous computing resources,the number of scientific workflow applications in heterogeneous systems is also increasing.Sometimes,the cloud resources are located in different geographically distributed regions.Therefore,allocating different heterogeneous cloud resources to such workflow tasks with satisfactory QoS requirements is quite challenging.Despite that,this kind of workflow application scheduling is critical for reducing the execution time of scientific applications,the rental cost of cloud users,and the electricity cost of cloud service providers.Most of the previous research work has mainly focused on optimizing the execution time cost and paid less attention to the energy consumption.To this end,this paper focuses on the energy consumption problem in heterogeneous virtualized cloud computing systems by focusing on energy-efficient task scheduling algorithms.The main work of the thesis includes:(1)The Energy and Performance-Efficient Task Scheduling(EPETS)algorithm is proposed to achieve good performance and reduce the total energy consumption while meeting deadlines.An EPETS algorithm includes initial scheduling and task reassignment scheduling algorithms.Initial scheduling requires task ordering,primary allocation,and execution slot allocation.In initial scheduling,we attempted to delegate maximum and lower slack time tasks to the fastest machines without considering energy optimization.The task reassignment algorithm helps minimize energy consumption while meeting deadlines by shifting tasks from a faster machine to a medium or slower machine.Simulation results show that the proposed EPETS algorithm performed better in energy consumption than current algorithms.This study considered scheduling a set of independent tasks with quality of service criteria(e.g.,deadline,different workload size,and execution priority)in heterogeneous cloud computing systems.However,it becomes more difficult for DAG-based workflow tasks on cloud resources in the GD-CDCs(Geographically Distributed Cloud Data Centers).With this in mind,we developed an effective method to solve DAG-based workflow task scheduling problems in GD-CDCs in the following contribution.(2)We developed a Deadline-constrained Energy-aware Workflow Scheduling(DEWS)algorithm to minimize electricity cost when scheduling DAG-based workflow tasks to GD-CDCs,in which electricity prices change according to the different electricity suppliers and time intervals.A DEWS includes workflow task sequencing and scheduling.The initial task scheduling sequence obtained by EDF,SSF,and SWF is not always the best solution to minimize the total electricity cost.Therefore,we propose a Variable Neighborhood Descent(VND)with combined neighborhood structures for task sequence adjustment and search data center.Each task is then assigned to the appropriate VM by the proposed VM search using the DVFS method.In this method,the VM frequency is dynamically adjusted based on DVFS to utilize the task slack time fully and further reduce the electricity costs for the service providers.We verified that the proposed DEWS outperforms the algorithms for the studied problem by comparing it with four benchmark algorithms.(3)The work done in the above two parts focused on independent or dependent task scheduling in heterogeneous virtualized cloud computing systems to optimize a single objective rather than multiple objectives.As cloud computing allows various user-provided applications to run in a virtualized cloud environment,many applications(tasks)are deadline-sensitive.When these applications are submitted to a virtualized cloud,they must be appropriately assigned to virtual machines to prevent deadline violations,shorten their completion time(makespan),and achieve energy efficiency.Given the conflicting nature of these two parameters,makespan,and energy consumption of the virtualized cloud infrastructure,minimizing both parameters simultaneously is quite challenging.We propose a Multi-objective Quantum-inspired Genetic Algorithm(MQGA)to address this issue.MQGA is inspired by quantum computing and classical genetic algorithms.It uses a qubit to represent the individual chromosome because it has better population diversity.It also uses a quantum rotation gate to refine the population to find better schedule solutions and avoid classical genetic operators.The study also proposes a new solution for search operators to explore the solution space fully.Experiments are conducted with different workloads to evaluate QoS,makespan,and energy consumption performance.Comparing the algorithms and discussing the results show that our proposed MQGA algorithm has better scheduling performance than the existing algorithms.(4)Finally,this work explores the problem of task scheduling in a hybrid cloud,which involves the following actions:(ⅰ)first try to run maximum and big tasks on the private cloud under defined deadlines,and(ⅱ)delegate the remaining unscheduled tasks to the public cloud,then(ⅲ)delegate all tasks to run on the private cloud or communication channels to run tasks on the public cloud to satisfy the task priority and deadline constraint requirements.At the same time,the total monetary cost is minimized.Deadline-constrained Cost-aware Workflow Scheduling(DCWS)heuristic is proposed,which consists of deadline division,task sequencing,and resource allocation.In this study,the idea of a sub-deadline is used to reduce the slack time of a task and find the most appropriate computational resource to reduce additional execution time and cost.Simulation results show that the proposed algorithm can achieve the highest cost reduction compared to existing algorithms. |
