Flow Characteristics Based Routing And Scheduling Technology Research In Data Center Networks

In recent years,the global data volume has been growing exponentially with the rise of the Internet era,and the data centers have sprung up all over the world.In China's 13th Five-Year Plan,some emerging industries such as big data and cloud computing have been listed as important strategic research directions.Meanwhile,data centers have become a comprehensive embodiment of the national informatization level,as data centers are providing infrastructure platform support for big data and cloud computing applications.As a vital part of the data centers,network is the key to intercommunication within one datacenter and among different data centers.Therefore,the performance of the data center network will largely determine the performance of the whole data center.With the service types increasing and the traffic patterns continuously evolving in data centers,data center networks are facing with challenges such as high congestion risks and weak job deadline guarantee capabilities.In order to make full use of the data center resources and provide satisfactory services for users,this thesis discusses how to effectively avoid data center network congestion and meet the deadlines by utilizing routing and flow scheduling methods.The traffic characteristic based congestion avoidance rerouting method for data center networks is proposed in this thesis,which can reduce the average packet loss rate and promote the average network throughput.Firstly,the characteristics of large and small flows in data center networks are perceived,and then the link criticality metric related to link load and delay performance is designed to represent the influence of large flows on the link.The higher one link's criticality is,the greater the influence of large flows on the link will be.Secondly,the rerouting path allocation problem is modeled with the objective of minimizing the maximal link criticality.Next,since this min-max rerouting path allocation optimization problem is NP-Hard,a heuristic rerouting algorithm based on the path criticality is proposed,which schedules large flows to the paths with lower criticality to alleviate the current congestions and avoid congestions on rerouting paths.Finally,the effectiveness of the proposed heuristic rerouting algorithm has been demonstrated by simulation results.Aimed to handle the deadline guarantee problem for multi-jobs with multi-stages and multi-flows,the job deadline guaranteed routing and scheduling method is proposed for data center networks.Firstly,based on the network flow constraints,a routing optimization model is formulated for multi-jobs with multi-stages and multi-flows,with the objective of maximizing the number of deadline guaranteed jobs.Secondly,a two-stage multi-flow routing algorithm is proposed.In the first stage,multi-jobs are scheduled in time dimension according to their deadline constraints;in the second stage,an iterative coflow joint routing algorithm is proposed,in which the coflow joint routing and scheduling problem was modelled as an integer programming problem and then relaxed to solve.In the iterative steps,at first the multi-flow paths are solved with fixed flow rates,then the maximum transmission rate of each flow is solved with determined flow paths and link capacity constraints,and those two steps are iteratively solved to obtain the final routed multi-flow paths with minimum job completion times.Finally,the simulation results prove that the algorithm proposed in this thesis can significantly increase the number of jobs that meet their deadlines.