Research On Load Balancing Technology In Data Centers

With the rapid development of cloud computing and big data processing technology,data center networks as cloud computing infrastructure have become a research hotspot.In the context of the rapid development of cloud computing,flexible resource sharing modes,frequent task collaboration between servers,and storage and interaction of massive amounts of data will put forward higher requirements for the management and control of data center networks.Traditional data center networks have been unable to have fast,flexible,ondemand,and flexible capabilities to meet the needs of future network development.With the help of centralized control,software-defined network technology can well meet the current development needs of data center networks.With the expansion of network scale and the sudden growth of traffic,software-defined data center networks will face two main problems,namely,unbalanced controller load and unbalanced traffic load.In response to these two problems,the existing research work is divided into controller-oriented load balancing technology and traffic-oriented load balancing technology.The controller-oriented load balancing technology realizes the balanced distribution of the controller load through the controller deployment optimization mechanism and the switch migration optimization mechanism.The traffic-oriented load balancing technology solves the problem of resource competition between mixed flows and topology asymmetry through the traffic optimization strategy and the load balancing strategy,so as to realize the balanced distribution of the traffic load.Combining the latest research progress of load balancing technology in data centers,this paper starts from the two perspectives of controller load balancing and traffic load balancing,with the goal of improving the utilization of data center network resources,focusing on controller deployment,switch migration,traffic optimization and load balancing Conduct research to solve the main problems that the existing load balancing technology affects the performance of the data center network,and achieve the design goals of high performance,robustness and scalability of the data center network.The main research results of this paper are as follows:1.Most existing controller deployment optimization mechanisms only consider optimizing one or a small number of performance indicators to implement controller deployment,and do not consider the dynamic variability of traffic.In the actual controller deployment process,the load balancing performance of the controller is low.In response to this problem,this paper proposes a multi-controller deployment optimization mechanism based on two-sided matching to achieve load balancing among multiple controllers.This mechanism converts the controller deployment problem into a two-sided matching problem,and considers optimization of multiple performance indicators that affect the controller deployment.In the actual operation process,when the controller is overloaded,the proposed switch migration algorithm is used to reduce the load of the controller.Compared with the existing research schemes,the proposed optimization mechanism improves the load balancing performance of 31.2% among multiple controllers on average.2.Most existing switch migration optimization mechanisms have low migration efficiency.For example,after the switch is migrated,the target controller may be overloaded,the load balancing performance among multiple controllers has not been significantly improved,and the switch migration will bring additional migration cost.In response to this problem,this paper proposes an highly-efficient switch migration optimization mechanism to achieve balanced distribution of controller load.This mechanism accurately measures the actual load of the controller by defining multiple evaluation indicators,including the controller's processing overhead for flow request messages,the overhead of routing rule formulation,and the overhead of state synchronization among controllers.During the switch migration process,the mechanism selects the controller with the largest remaining resources as the target controller,thereby improving the load balancing performance of the controller.In order to improve the efficiency of switch migration,this mechanism selects the optimal switch from the set of switches associated with the overloaded controller as the switch to be migrated by minimizing the switch migration cost.Compared with the existing scheme,this mechanism improves the load balancing performance of 45.6% among multiple controllers,and reduces the migration cost of 43.5%.3.Since most of the existing traffic optimization strategies for resource competition between mixed flows have the problem of low network performance and scalability,this paper proposes a high-performance and scalable traffic optimization strategy.This strategy combines the advantages of centralized and distributed mechanisms at the same time,through priority scheduling to ensure that the mice flow is preferentially transmitted,and the end host adopts the congestion control based on redundant coding to adjust the transmission rate of the mice flow,thereby reducing the latency of mice flows.This strategy schedules elephant flows by deploying a marginal cost-aware dynamic flow scheduling on a centralized controller to improve their throughput.In the actual execution process,the centralized controller only processes the elephant flow,thus achieving better scalability.Compared with the existing traffic optimization strategy,this strategy can reduce the Flow Completion Time(FCT)of the mice flow by an average of 40%,and increase the throughput of the elephant flow by an average of35%.4.Since most of the existing load balancing strategies for topology asymmetry have the problem of low network robustness and scalability,this paper proposes a robust and scalable load balancing strategy.This strategy routes flowcell to the path with the least congestion through global congestion awareness,thereby achieving better network robustness.In order to collect and store the congestion information of the entire network,this strategy proposes a distributed control structure,and monitors and collects the congestion information of the entire network through multiple controllers,thereby achieving better network scalability.Compared with the existing research scheme,under asymmetric topology,this strategy reduces the FCT of mice flows by19.8% on average,and improves the throughput of elephant flows by 15.5%.