Research On Fine-grained Time Load Balancing Method For Data Center Network

With the rapid development of data center network transmission technology,the network transmission rate has grown exponentially,and the propagation delay has also been continuously reduced.At the same time,the traffic in the network also exhibits the characteristics of burst transmission in a very short time.Generally,the traffic with such characteristics is called Microburst^[1,2,3].In view of this kind of traffic,most of the existing load balancing methods cannot effectively offload and forward it,so network anomalies such as excessive switch buffer occupancy,link congestion,and packet loss are prone to occur during traffic transmission.Therefore,in order to cope with the balancing and forwarding of fine-grained time burst traffic,this paper improves the existing load balancing methods and designs two new methods to improve the balancing effect of Microburst traffic,reduce congestion probability and enhance network transmission performance.The Drill^[4]method is a load balancing method based on the packet granularity.This method achieves the ideal balancing effect of the ESF?Equal Split Fluid?model by realizing the conditions of the ESF model.Therefore,this method can effectively deal with the balancing and forwarding of Microburst traffic.However,the balancing effect of the Drill method depends on the symmetry of the network.The worse the network symmetry,the closer the actual balancing effect of the Drill method is to the ECMP?Equal-Cost-Multi-Path?^[5,6]method.Therefore,in order to cope with the performance degradation of the Drill method in asymmetric networks,this paper proposes an improved load balancing method.The improved method is based on the packet balancing granularity and borrows the Drill method's design idea to simulate the ESF model to achieve a better balancing effect.At the same time,for the load balancing in asymmetric networks,the improved method adds a network congestion status monitoring mechanism.It uses the obtained network congestion information to make forwarding decisions to reduce the impact of link congestion on traffic transmission in an asymmetric network.Thus it can improve the balancing effect of Microburst traffic and enhance the overall network transmission performance.Since the flowlet^[7]balancing granularity can not only ensure a good balancing effect,but also control the packet reordering to some extent,it has been adopted by more and more load balancing methods,such as the LetFlow^[8]method.However,the excessively passive flowlet segmentation mechanism makes the LetFlow method unable to effectively balance and forward Microburst traffic.Therefore,this paper designs an improved method based on the flowlet granularity on the basis of the LetFlow method.The improved method inherits many advantages of the LetFlow method,and adds a flow monitoring and active segmentation mechanism to the defect that the flowlet segmentation mechanism is too passive.It can actively split the monitored burst traffic,so that the burst traffic can be divided and forwarded in the network.Thus,the method improves the balancing effect and reduces the impact of burst traffic on network transmission performance.In the end,this paper also verifies the improved effects of the two improved methods through simulation and comparison experiments.