Efficient Coflow Scheduling For Time Optimization In Data Center Networks

In recent years,the network has become a bottleneck limiting the development of data centers.Reducing data transmission time and deadline guarantees are two important performance goals in the data transmission process.Existing prior knowledge is unknown,and traffic characteristics are not fully utilized when making scheduling decisions.There is still room for optimization in terms of the average completion time of the Coflow.Therefore,this thesis designs a multi threshold prior knowledge unknown Coflow scheduling scheme based on traffic characteristics to further reduce the average completion time of Coflow.In addition,existing studies either focus on reducing the average completion time of Coflow or focus on deadline guarantees,but in actual network environments,simply focusing on one performance goal can harm another performance goal.Based on this,this thesis studies multi-performance target traffic scheduling in data center networks,aiming to maximize the number of Coflows that meet deadlines while minimizing the impact on the average completion time of Coflows.In summary,this article mainly studies the following two aspects:In the first work of this thesis,in order to further reduce the average completion time of Coflow in scheduling schemes with unknown prior knowledge,a multi threshold classification scheduling algorithm A-SNCF is designed using the traffic characteristics in data center networks.When the coflow arrives,A-SNCF assigns different initial priorities to the coflow based on its width,avoiding the obstruction of small coflows to large ones.During transmission,A-SNCF divides Coflows into two categories: wide and narrow,and adopts different priority adjustment methods for the two types of Coflows.Simulation experiments show that A-SNCF can significantly reduce the average completion time of Coflow compared to classic scheduling schemes with prior knowledge unknown such as MCS and Aalo,with good performance.The second task of this article is to design a scheduling algorithm S-DLCF for traffic in data center networks,with the goal of minimizing the impact on the average completion time of Coflows and maximizing the number of Coflows that meet deadlines.Due to the differences in traffic size and deadline in data center networks,and the size of the Coflow and its subflow information are unknown before transmission is completed,this article uses traffic deadline to optimize traffic transmission based on the prediction of the Coflow size and priority allocation using traffic characteristics.Specifically,based on the traffic characteristics,the initial priority is assigned to the Coflow,and different priority adjustment methods are used for different types of traffic,And sort the Coflows in the same priority queue according to the deadline.Simulation experiments show that compared to scheduling algorithms such as Aalo and FIFO,the proportion of Coflows completed within the deadline in S-DLCF is higher,and the impact on the average completion time of Coflows is minimal.S-DLCF has good performance in deadline guarantee and minimizing the average completion time of the coflow.In general,in scenarios where prior knowledge is unknown,this thesis first studies scheduling with the goal of minimizing the average completion time of the Coflow,and designs and proposes a scheduling algorithm A-SNCF.Then,based on this,this thesis utilizes the deadline for Coflow,designs and proposes a scheduling algorithm S-DLCF,which improves the proportion of Coflow completed within the deadline while ensuring the average completion time of the Coflow.Simulation experiments show that both A-SNCF and S-DLCF have good performance.