Research On Key Technologies Of SLA-Based Resource Scheduling In Cloud Computing Envi-Ronment

With the rapid development of Internet technology and the widespread application of virtual-ization technology, the cloud computing, as a service of supporting on-demand and pay-as-you-go, has won the acceptance both in academia circle and industry world.The resource management is a key element to influence the properties of cloud computing system. The cloud computing, which is supporting by large-scale cluster computing systems, needs effective resource management techniques to protect the user application reliability and availability of resources, in order to achieve the security of cloud computing.Due to the utility characteristics of cloud, the cloud environment for users and resources is constantly changing. It becomes an urgent problem in the cloud computing environment that how to plan and schedule and adjust the cloud resources in this dynamic environment in order to match the needs of users. A Service Level Agreement is the specification to guarantee expectations and re-sponsibilities for the service quality, price and other factors between users and providers. The SLA constraints can improve users’ satisfaction and resource utilization as much as possible, so as to promote the user and cloud service providers to join the cloud. All above this promote the research-ers to innovate the scheduling technology of cloud resources with the constraint conditions of the goal of service level agreements. Therefore, this thesis focuses on the cloud resources scheduling under constraint conditions of SLA, mainly includes the following four aspects:1. This thesis studies the architecture of cloud resources’scheduling under SLA constraints to solve the problem that there’s no clear limit between resources services and value-added services in cloud environment by stripping the routine of SLAs’ negotiation and production and maintenance and resource scheduling to the broker. A novel architecture of resource scheduling which is re-stricted by SLA is proposed and cloud broker is designed as the core unit of this architecture. This architecture realizes on-demand and flexibility and extensibility of resource providing in cloud en-vironment. Further, a buffer which is diffserv crossbar structure similarly as switch buffer in net-works is fixed in the broker to reduce the task loss rate and tasks’ waiting queue. The results of sim-ulation show that the resource scheduling architecture under the SLAs’ restriction can effectively ensure the application reliability of users and the availability of resources.2. This thesis investiagtes the resources selections under the demand of users’to solve the complexity problem that the available resources can’t match the applications’offer. The broker split the application into a tasks’queue with workflow and aimed at these tasks request resources respec-tively. Based on the resource selection process same as the SLA matching process, a fuzzy evalua-tion method is proposed to satisfy tasks’demand by matching the SLA between tasks and resources. The method describes distance between brokers and users by introduce weights of SLA, express the content of SLA that need to be matched by SLAs’ value, select the best match resource by mutual evaluation between brokers, optimize the selection algorithm by workflow relationships, in order to improve the reliability of the application in a cloud environment. The results of simulation show that the method can reduce the overall cost of application and improve the reliability of resources.3. This thesis considers the virtual machine scheduling on the cloud infrastructure to solve the problem of how to use the resources with the highest possibility. From the perspective of infra-structure, resource scheduling is the same as virtual machine scheduling, and the methods is related to the infrastructure efficiency and infrastructure service provider’s earnings. A method with two levels of task management and resource matching is proposed to improve the utility of cloud re-sources, reduce the cost of the virtual machine scheduling, satisfy the SLA constraints of tasks, so as to ensure the availability of cloud resources. Task management model fixed in broker manages the virtual machines into three steps as prediction, feedback and judge. The prediction model is multilayer curve fitting in order to forecast the possible virtual machines requirements. The feed-back model will be to calculate the distance between feedback SLA and resources’SLA threshold to determine whether additional virtual machines requirements are produced. At last, judge step syn-thesizes the result of prediction and feedback, and then using moving average to determine the de-mand of virtual machine. At SLA matching level, we use multifactor matching method of supply and demand to find an optimal physical machine to satisfy the tasks’resource requirements. The results of simulation show that the virtual machine scheduling under the restriction of the SLA can effectively improve the utilization rate of resources and ensure the availability and reliability of cloud resources.4. This thesis discusses the scheduling method at hybrid cloud environment to satisfy the flexibility of large-scale resource sharing and real-time demand of applications. By the dynamic and elastic characteristics of cloud, the broker should rent the extra services to satisfy the dynamic change of users’demand when the current cloud’s resources can’t satisfy the demand of the current situation. A dynamic programming method is proposed to improve internal cloud resource utiliza-tion and minimize external cloud resource rent expense to solve the cloud resource scheduling un-der SLA constraints. The results of simulation show that the scheme can maximize resources utili-zation of the internal cloud and minimize the rent cost of external cloud resources, and ensure the reliability of the application and the availability of resources.