The Scheduling Methods Based On The Task Features And Resource Constraints In Cloud Computing

Cloud computing is a commercialization service model. It allocates resources on-demand and develops rapidly. It is very important to research the methods of allocating resources according to requirements submitted by users. However, there are some problems such as the dynamic, ultra large scale of cloud resources, the high energy consumption, the diversity of the demand for resources of tasks and diversification constraints of tasks. These problems seriously affect the scheduling efficient even the QoS in cloud computing, and become the factors restricting the development of cloud computing. Therefore, if some scheduling optimization methods can consider related characteristics of the task and resource, it will improve scheduling efficiency and quality of service, even more conducive to the development of cloud computing.Firstly, this paper surveys comprehensively the theme – the scheduling issues in cloud computing. It analyzes the definitions, objectives and features of the scheduling in cloud computing. And it discusses and summarizes the current status of scheduling from three principles of the performance, quality of service and economic. Especially, it in-depth researches the current cloud computing platform and their scheduling methods. At last, it analyzes some problems which existing the current researches on scheduling in cloud computing, and addresses the further work.(1) In order to precisely describe and reflect the dynamic of resources, the diversity of tasks, the the multi-constraint and other characteristics of resources and tasks, this paper proposes a resource-task model based on entropy optimization.It analyzes and mines the characteristics of tasks and resources. Then it pretreats resources by dynamic clustering through some dynamic evaluation indicators expressing the computing power of resources. Aiming at the diversity of tasks, it classificies tasks by the requirments of tasks to resources. Then entropy optimization model is used to reflecting the characteristics of the task resource demand characteristics and the task itself based on resources pretreatment and tasks classification. The model is versatile and can learn from the other cloud computing applications according to the application characteristics.(2) In order to solve these problems of the dynamic, the ultra-large-scale and the high energy consumption, this paper proposes a dynamic scheduling optimization methods for energy-aware-- STDWEM(Self-adaptive Threshold Dynamic Weighted Evaluation and Scheduling Mothod).Aiming for the dynamic and ultra-large scale issues, this paper proposes a dynamic scheduling optimization method for energy-aware based on the dynamic load evaluation in the resource-task model of the third chapter. This method migrate the overload resources to balance load and improve resource utilization; at the same time, it releases the very idle resources to save energy. Thus this method achieves the multi-objective integration and optimization of resources. This method takes some task migration coefficient and adaptive thresholds when migrating resources, and it also considers the load state of resources when allocating tasks. Meanwhile, the scheduling method further comprises an energy evaluation model that can quantitatively describe the energy consumption before and after the integration of resources. In order to verify the validity of the scheduling method, experiments verified the effectiveness of this method on scheduling performance, dynamic adaptability, energy and so on. Experimental results show that it is conducive to improve the scheduling efficiency the dynamic clustering of resource and considering resource load when scheduling task. This method can significantly reduce energy consumption, nearly 31.5% of the maximum energy savings comparing with the similar scheduling mehods which also use the resource evaluation and consider energy consumption. Moreover, this method has advantage on response time and system utilization. Especially, it shows good stability and adaptability when resources are dynamically joining or leaving.(3) In order to solve the dynamic of resources and the requirement diversity of tasks for resources, this paper proposes an interlacing multi-queue scheduling method--MIPSM(Multi-queue Interlacing Peak Scheduling Method).First, the diversity tasks are divided into three tasks queue--CPU intensive, I / O-intensive and memory-intensive tasks by the clustering method in the resource-task model. The power factors of the CPU, I/O and memory can be adjusted to reflect the different emphasis of different application tasks when classifying. On the basis of the tasks classification, tasks are allocated to resources by the state of load resources based on differences in task demand for resources, resources are sorted three queues by the CPU, I / O and memory load. Then tasks are scheduling to those resources whose load is light. Thus tasks are scheduled according to resource load difference and it brings very good scheduling efficiency. Some simulation and real workload experiments are designed to verify the effect of task classification and scheduling methods. The results show that task classification significantly helps to improve scheduling efficiency. The scheduling method in this paper has a very good advantage on response time, resource utilization, the deadline violation rate. Especially, it shows a greater advantage with the increasing number of tasks.(4) Aiming at the problems of the diversity constraints of tasks from the user and resources in cloud computing, this paper proposes a multi-objective optimization scheduling method based on the diversity constraints of tasks--MOSACO(Multi-objective Optimization Scheduling Method Based on Ant Colony Optimization Algorithm).This method first proposes two single-objective optimization strategies--Time-First and Cost-First. Time-First and Cost-First are respectively priority to the time and cost based on diversity constraints of the task deadline and budget cost. In order to maximize the interests of user QoS and resource provider their objective functions are as constrains of entropy optimization model in resource-task model. Then the entropy optimization model combines these two scheduling methods to establish a multi-objective optimization scheduling model which is solved by ants colony algorithm. Experiment results show that, compared to other similar methods, the multi-objective optimization methods has advantage on the completion time, cost, the deadline violate rate and resource utilization. Cost-First exhibites advantages in terms of cost. Time-First is also a great advantage on the completion time. All these demonstrate the effectiveness of the proposed scheduling method in this paper.