Research And Implementation Of Load Balancing Algorithm For Data Center Network

Today,with the rapid development of the Internet,network technology has entered a new era,and a variety of applications emerge endlessly.Behind the development of this series of services is the explosive growth of data volume.The traditional processing technology has been unable to adapt to the development of the times,resulting in the data centers,which shows great advantages in data computing,transmission,storage and other aspects.More and more enterprises and organizations are building data centers to deploy their own businesses,and the number of data centers is increasing.Nowadays,cloud computing,5G mobile communication,artificial intelligence,Internet of Things and other industries are developing rapidly from scratch.Data centers play an increasingly important role in supporting multimedia content,e-commerce and other cloud services.Today's data centers typically contain hundreds of servers with large scale.Each server node is configured with multiple paths.Each link uses high bandwidth to ensure high-speed traffic transmission.At the same time,with the increasing traffic,the transmission performance of the network determines the user's experience,and the load balancing technology for the data flow is the key to improve the network performance,which has certain value and significance.The first problem to be solved in an efficient load balancing scheme is the data flow routing.At present,in the research of data flow scheduling,the technical scheme based on packet granularity can make full use of the multi-path characteristics of data center network.Each packet in the flow,as a separate decision-making individual,can be allocated to all reachable paths and transmitted through multiple links,rather than selecting one of the paths.By analyzing the existing technical solutions,this paper proposes a packet routing algorithm based on six tuples.Six tuples are based on five tuples,and an additional value is introduced.These six-information data can distinguish different packets in the data stream.In the process of packet transmission,the switch uses the hash algorithm based on six tuples to select the forwarding path,so that the packets in the same stream can be transmitted along different paths,realizing the technology of taking packets as the basic scheduling unit,and reducing the completion time of the data stream.The algorithm is simple and easy to implement,without adding additional overhead.In the research of load balancing technology,timely understanding of the congestion occurring in the network and the congestion degree of each link node,making full use of the topological characteristics of the data center network,and dynamically adjusting the routing distribution of data packets on different paths,which can effectively alleviate the congestion.This will reduce the accumulation of data packets and reduce the occurrence of packet loss.The centralized solution relies on the centralized controller to obtain the state of the whole network,while the distributed solution relies on the switch or each host to judge the path status.Based on the distributed design idea,this paper proposes a load balancing scheme based on the probe packets.Before sending the data packet,the probe packets are first sent to test the transmission capability of different paths.According to the arrival order of the probe packets,after understanding the performance of each link,the data packets are sent based on the detection results.The probe packet adopts a six-tuple method.By assigning the sixth information value of the probe packet to the corresponding data packet,the transmission path of the data packet and the probe packet are consistent.Simulation experiments show that the algorithm can balance the traffic load of each path,reduce transmission delay,packet loss rate,etc.,and improve the performance of data center network.