Research And Implementation Of Task-level Flow Scheduling Mechanism In Data Center Virtualization Environment

With the continuous maturity of cloud computing technologies and the abundance of cloud services,more and more applications are deployed in the cloud data center virtualization environment.The performance of these applications is closely related to the network transmission because of the distributed features.The network transmission requirements can often be expressed as tasks and each task consists of several distributed flows.According to the network transmission performance requirements,tasks can be divided into delay-sensitive tasks,such Web service tasks,and bandwidth-sensitive tasks,such as MapReduce tasks.The coexistence of the heterogeneous and mixed tasks poses great challenges to data center flow scheduling.Firstly,different types of tasks compete for shared network resources,resulting in low performance of network transmission.Secondly,bandwidth-sensitive tasks with different utility features get resources without discrimination,resulting in low resource utility.The existing work does not consider the heterogeneous task characteristics hence cannot deal with the above challenges.Therefore,in this paper,the task-level flow scheduling problem is studied and the work is carried out from the following aspects.Firstly,in order to solve the problem of low performance of network transmission due to task competition sharing network resources,the task type aware flow queue management mechanism is studied.The mechanism provides network transmission performance for different types of tasks based on task type awareness.Meanwhile,in view of the dynamic feature of application network load intensity,it adjusts the weight of different types of tasks,so as to better meet the demand of different types of applications.Secondly,in order to solve the problem of low resource utility caused by the indifferent resource allocation with different utility features,the global flow scheduling mechanism oriented to heterogeneous time sensitivity is studied.The mechanism conducts a unified scheduling for different flows in the same task,and uses the utility function to model the sensitive degree of different tasks on completion time,ensuring the utility fairness among tasks while improving the total bandwidth allocation utility.Finally,this paper designs and implements a prototype system in SEU Cloud platform,applying the theoretical research results into practice.The experimental results show that the proposed task-level flow scheduling mechanism for can effectively promote the network transmission performance.Meanwhile,it ensures the utility fairness between tasks and improves the system total bandwidth allocation utility.