Research On Techniques Of Flow Scheduling And Request Allocation In Data Centers

With the development of the Internet and hardware devices,data centers have evolved into three generations as an infrastructure for deploying various applications:single data center,remote multi-data centers,and small scale edge data centers.Nowadays data centers are not only growing in size,but their granularity is also becoming finer.Data centers meet many types of user needs,there are still more problems need to be solved.This thesis analyzes and compares four difficult points in the data center with limited network resources,diverse task requirements,huge data volume and plenty of energy consumption from three levels of single data center,multi-data center and small edge data center.Comparing with current works,this thesis studies two types of problems traffic scheduling and request allocation.The detailed research content and contributions are summarized as follows:In terms of data center network,this thesis studies deadline aware flow scheduling problem.There are multiple types of applications,one user request requires multiple flows to work togeth-er,and then the one determines the completion time of a task is the slowest one.At the same time,the some applications deployed in the data center are delay-sensitive,and the transmission completion time of the flow has a great influence on the performance of the application.To this end,this thesis proposes a task-level flow scheduling method based on deadline.Different types of applications are divided into different priorities,and the priority definition of the flow level inherits the attributes of the application.At the flow level,the flow adjusts its priority based on the deadline.This thesis establishes a model of revenue generated by flow scheduling,and based on this,designs a flow scheduling algorithm.Simulation experiments show that the proposed method compares the current flow scheduling methods with deadlines while improving the task completion and reducing the overall task completion time.In terms of inter-datacenter network,this thesis focuses on traffic allocation and request allocation issues with user performance experiences and energy consumption.As for perfor-mance experience,more and more Internet companies choose to build their own data centers in order to achieve the goal of globalized and guaranteed services quality.We apply the geo-graphical distributed data centers to provide multi-path selection for data transmission among data centers,while providing guaranteed service performance.However,the network status is complex and changeable,and burst traffic has a great impact on these characteristics.To this end,this thesis proposes a network congestion-based traffic allocation method,whose goal is to reduce the impact of congestion on traffic transmission while to maximize the overall profit.In order to solve the above problem,we use matrix transformation and mathematical derivations.Simulation results show that the proposed method can effectively improve the benefits of traffic allocation and reducing the damage caused by network congestion.In addition,as for energy consumption problem,we find that geographically distributed data centers are an essential infrastructure for providing reliable services to customers.Accord-ing to current researches,energy consumption occupies a large amount of cost of data center providers,so reducing energy consumption and ensuring reliable cloud service performance are significant for these providers.We find that user requirements are redundant over multiple data centers,an energy efficient request allocation method is proposed in this thesis.We jointly con-sider energy consumption and the quality of service of network rather than solely one factor on network or energy consumption.In order to achieve the reliability of performance,we propose a method of resource evaluation method.We use a smooth technology to solve the proposed optimization problem.The experimental results show that our method can effectively reduce energy consumption while ensuring service quality.In terms of edge system,this thesis mainly focuses on the research of user requests allo-cation problem.With the development of the Internet of Things and cloud computing,edge computing is emerging.Computing capacity is deployed on terminal devices or network nodes,which can help mobile devices have close computing capabilities when data is not uploaded to the cloud.For mobile users,computing performance and power consumption are critical.Com-puting task offloading to an edge system,while most of them have strict time requirements.To this end,we propose an offloading method aims to reduce energy consumption and can guar-antee the performance of computing tasks.The method calculates the partial offloading of task to the edge cloud server according to the requirements of the task and the energy consumption reservation of the mobile user,we use KKT conditions and Lagrange multipliers to solve this problem.Experiments shows that our method can reduce energy consumption without damaging the quality of task completion time.