Research On Autonomic Scheduling Technology In Cloud Computing

Cloud computing is the product of information technology revolution and an extra large-scale distributed computing model.In the cloud computing environment,massive resources are dispersed around the world.The cloud computing system is dealing with massive data and application tasks,with a high degree of dynamic.Because the cloud computing system is featured by large resource scale,heterogeneity and diversity,wide user base,different types of tasks and different QoS target constraints,etc.,it needs to frequently schedule and manage cloud resources and application tasks.Cloud computing,as a commercial computing model,aims to achieve resource sharing and collaboration and meet the user's service requests and service benefits of cloud service providers.Therefore,how to allocate cloud resources reasonably and manage and schedule the massive application tasks effectively,and how to reduce the scheduling cost of the users and improve the service revenue of the cloud service providers under the premise of ensuring the load balancing of the cloud computing system and improving the utilization rate of the cloud resources,are research hotspots in the task scheduling field of the cloud computing environment.In essence,the task scheduling in the cloud computing environment is a scheduling strategy which needs to establish a mapping relation between the application task and the computing resource.The purpose is to realize the rational allocation of computing resources and the efficient execution of the application tasks.Early traditional distributed computing system,grid computing system,etc.,measure the efficiency of a distribution mapping relation with the system throughput and task completion time and other performance indicators.However,in the cloud computing environment,the task scheduling problem is often more complex.First,the application task request in the cloud computing environment is often decentralized,and has a large scale.The traditional centralized resource management system is no longer applicable,and the cloud computing system must schedule and manage the tasks and the resources in a distributed and parallel mode;secondly,as cloud resources are usually from different cloud service providers,each of whom often offer heterogeneous cloud resources,how to combine a variety of heterogeneous resources into a complete service and provide them to the users are very complex;Moreover,as the cloud computing system has a dynamic scalable nature,the corresponding task scheduling must also meet the scalability and adaptability;then,as the cloud computing is a business model,cloud service providers firstly consider how to improve their own resources utilization,and get as much service revenue as possible.Thus,it's necessary to consider the overall load balancing of cloud computing systems at the moment of task scheduling.Finally,from the user's point of view,whether the services provided by the cloud service providers meet QoS multi-objective requirements of the user's task scheduling is the issue that must be considered by the task scheduling strategy.Therefore,considering the problems arising from task scheduling in the current cloud computing environment,the thesis makes a detailed and systematic study on the task scheduling technology in the cloud environment.The main contents include four aspects: QoS multi-objective optimization of user's task scheduling in cloud environment,cloud resource utilization and cloud computing system load balancing,implementation efficiency of cloud computing system task scheduling strategy,and user's task scheduling cost and service revenue of cloud service provider.Based on the deep research on these problems,the thesis designs a new set of autonomous scheduling architecture under the cloud computing environment,and gives detailed algorithms and experimental results for the important functions.The main research contents and innovation points of the thesis are divided into the following aspects:1)On the basis of analyzing and studying the problems and objectives requirements of task scheduling in cloud computing environment,the thesis proposes an autonomous scheduling architecture under cloud computing environment.In the autonomous scheduling architecture,the system creates a cloud broker for each user requesting the service from the cloud computing system,and the cloud broker has the whole authority to make the decision on behalf of its users.The functions and algorithms,such as analysis,evaluation and game,etc.,are also designed for each could broker.In this paper,a two-tier topdown autonomous scheduling model is designed by classifying functions of cloud broker.In addition,a cloud broker interaction model is designed for the negotiation between users and cloud service providers.The cloud broker selects the SLA protocol that conforms to the user's QoS target constraint via a series of constraint rules and target rules,and evaluates the SAL protocol.Secondly,the cloud broker game model is designed to solve the issue related to the resource conflict caused by user's competition for key resources.The cloud broker introduces the global revenue function of the system into the profit function of the individual user.When meeting the QoS multi-objective optimization requirement of the user,the cloud broker adopts the mode of autonomous competition game to maximize the overall benefit of the whole system.The cloud broker realizes the autonomic scheduling of the user task by means of negotiation and autonomous game,improves the execution efficiency of the task scheduling strategy,reduces the scheduling cost of the user,and improves the service revenue of the cloud service provider.2)Targeting at the issue about how the user judges whether the cloud service provider fulfills its commitments as set forth in SLA protocol in the cloud computing environment,the thesis presents a SLA protocol scheduling evaluation algorithm based on the star structure.The algorithm abstracts the resources promised by the cloud service provider in the SLA protocol into the standard virtual resource through the cloud broker,and abstracts the mapping relation between the user task and the virtual resource as the star topology cloud computing system.The quality of service that the cloud service provider is committed to is elevated by evaluating the performance and reliability of the cloud computing system.The users use the quality of service committed in SLA protocol as a criterion to determine whether the commitments made by the cloud service provider in the SLA protocol meet the needs of users.3)A multi-objective optimization scheduling algorithm based on NSGA-II is proposed for QoS multi-objective optimization of user task scheduling.In the cloud computing environment,on the one hand,it is still very difficult for users to choose a scheduling strategy that meets their QoS multi-objective optimization requirements among a set of valid SLA protocols.User application tasks can sign SLA protocols with multiple cloud service providers,whose SLA protocols are combined freely to often obtain an exponential figure.On the other hand,the QoS objectives of the users are often conflicting.The algorithm proposed in the thesis provides a fast and efficient selection mechanism for users,and the NSGA-II algorithm is used to perform parallel search on multiple targets.Compared with the traditional method of transforming multiple targets into single target,the algorithm proposed in the thesis can effectively choose a reasonable optimal SLA protocol scheduling strategy for users,when satisfying their QoS multi-objective optimization requirements.4)In the case where the SLA protocol has been signed,the user task chooses different cloud resources in the SLA protocol to constitute different task scheduling strategies.A scheduling evaluation algorithm based on virtual tree structure is proposed to determine which scheduling strategies meet the QoS multi-objective optimization requirements of users.In this paper,the task and specific cloud resources are modeled as virtual tree structures.The thesis adopts the minimum spanning tree algorithm to mathematically model the cloud resource and the transmission channel.In case a task is assigned to multiple cloud resources,the whole task can be completed,as long as one minimum spanning tree do not fail.In case several tasks share the same resource,the algorithm adds new branches and leaf nodes to the cloud resource node,and achieves the equal resource sharing for parallel tasks and exclusive resource sharing for serial tasks.Therefore,in the tree structure,the user task and the cloud resource is a dynamic and flexible mapping mode.The proposed algorithm can quickly and accurately evaluate any task scheduling strategy.As the feedback parameter of the user utility function in the game process of cloud broker as set forth in Chapter 6,the proposed algorithm makes each user get more efficient scheduling strategy and improve the utilization rate of cloud resources.5)Targeting at the problem that a large number of users submit task scheduling request at the same time and multiple users compete in key resources,the thesis puts forward a multi-user autonomous scheduling algorithm based on game theory.The algorithm assumes that each user has its own preferences and selfishness,and always tends to maximize its own interests.Each user will only choose its own most favorable resources,regardless of the impact on other users.As a result of the user's individual rationality,the prisoner's dilemma is formed,resulting in a resource allocation conflict.For example,most users choose the same resource,which results in load unbalance of the system.By means of the process of autonomous game between cloud brokers,the algorithm integrates the global utility of cloud computing system into the utility goal of each user,and designs a global incentive factor by designing the interaction rules between cloud brokers.Thus,the cloud computing system is allowed to obtain the global optimal revenue,when each cloud broker realizes their utility objectives.The independent game algorithm proposed in the thesis does not only enable each user to find the optimal resource tuning strategy in a competitive environment,but reduces the user's scheduling costs and improves the overall revenue of cloud service providers.