Research On Coflow Scheduling Mechanism And Simulation Platform Of SDN Based Data Center Network

This thesis focuses on the problem of minimizing the average completion time of the parallel computing service collaborative data stream in the data center network.Nowadays,there are a large number of parallel computing applications in data centers.Cooperative data traffic generated by such applications is usually described by the Coflow model.Reasonable Coflow traffic scheduling can reduce the average Coflow time.The study of minimizing Coflow's completion time is usually divided into two kinds of scenarios,which are known prior knowledge and prior knowledge unknown scenes.Taking into account the practicality,this article will focus on the prior knowledge unknown scene.At present,the research of the Coflow scheduling mechanism under this scenario is mostly based on the Aalo framework.This framework only uses Coflow size estimation to perform Coflow scheduling.In this thesis,Coflow scheduling scenarios are studied and analyzed.The scheduling mechanism of Aalo is not conducive to minimizing the average Coflow time in some scenarios.Therefore,this thesis focuses on the parallel computing application scenario of data center network with unknown prior knowledge and proposes a Coflow scheduling mechanism CSPAT based on port aggregation traffic.Detailed description of the mechanism,characteristics research and performance simulation of the mechanism are provided.Aiming at the problems existing in CSPAT,an improved solution CSPAT-A was proposed.In addition,this article also introduced in detail the "Coflow intelligent simulation system" based on the NS3 simulation platform,which is used for simulation verification of the Coflow scheduling mechanism.The work results of this article are now described in detail as follows:1.For a priori unknown scenario,a Coflow scheduling mechanism CSPAT based on port aggregate traffic is proposed.The main idea of the CSPAT scheduling mechanism is that the controller periodically calculates the estimated value of the aggregated traffic of the Coflow according to the estimated value of each subflow of the local Coflow reported by the host side,and formulates the Coflow priority according to the size of the estimated value of the aggregated traffic of the port.Level,each host and switch perform WFQ scheduling according to the Coflow priority.The CSPAT scheduling framework proposes two improvements on the basis of the Aalo scheduling framework:(1)the Coflow size estimation method has been improved;(2)the Coflow priority partitioning has been changed to Coflow's port aggregation traffic size instead of Coflow.size.Simulation results show that CSPAT has a 3% average optimization effect compared to Aalo in minimizing the average Coflow completion time.Its improved solution CSPAT-A has about 2% performance improvement over CSPAT.Both methods can be easily deployed on existing data center network nodes,which can effectively reduce the average completion time of Coflow in the communication phase of parallel computing applications.In order to carry out simulation verification of CSPAT scheduling mechanism,this thesis designs and implements Coflow intelligent simulation system based on NS3 simulation platform.The simulation system includes three parts: user interface,Web background,and NS3 simulation platform.Among them,the NS3 simulation platform is the result of transformation and development based on the NS3 source code,which mainly includes the realization of the controller module,the host module,and the switch module.The simulation system can support the verification of SDN-related theoretical algorithms,and provide a user-friendly graphical interface to complete a series of operations such as topology definition,Coflow flow injection,and simulation program call,and visualize the simulation results.