A Controllable Random Load Balancing Algorithm For Data Center Networks

With the development of network techniques and requirements of services,the deployment and research of data center networks have drawn growing concerns.Data center networks provide multiple high-speed links for services,which could transmit a large amount of data within a short period of time.In order to make full use of the multi-path feature of the network,a load balancing algorithm is needed to schedule flows in data centers.However,the characteristics of traffic in data centers are much different from those in traditional networks,so the load balancing algorithms designed for traditional networks cannot be directly deployed in the data center network.New algorithms should be designed for the data center network with taking its traffic features into consideration.The recent state-of-art load balancing algorithm for data centers leverage packets to carry congestion information to the sender,and then the sender assigns flows to the least congested path accordingly.This scheme performs suboptimally since its transmission of congestion information is not accurate and prompt enough.Furthermore,selecting least congested path would schedule concurrent flows to the same path,which will lead to a local optimal assignment.Based on the flaws of current load balancing algorithms,we proposed a controllable random load balancing algorithm for data center networks.By sending probes at the edge switch,the algorithm could collect and transmit congestion information in real time.It also introduces randomness to deal with concurrent flows when selecting paths.Simulation results show that the algorithm could solve the problem of suboptimum and improve the performance of data centers.