Research On Energy-saving Traffic Allocation In Fat-tree Networks

With the increasingly strained global power resources,the number of data centers and the power consumption of exchange equipment have grown.The power consumption of switching equipment in a large-scale data center is astonishing.However,the switching equipment which is under the long-term idle state has a low utilization rate,but the power consumption of an idle switch is almost the same as the power consumption of a full loaded switch.Therefore,the research on energy-saving strategies for switching equipment in data centers has attracted attention.This thesis aims at reducing the network energy consumption.With the coexistence of unicast and multicast mixed services in the fat-tree network structure,two network energy-saving algorithms are proposed.The energy-saving algorithm which only aims at consumption in network is not friendly to the performance of a network.Most packets will be out-of-order when reaching at the destination switch.Therefore,a load-sensing energy saving algorithm which is based on fat tree topology is proposed,and the routing algorithm can make packets arrive the destination switch approximately without disorder.The main tasks of this part are as flows:Different from the structure of virtual output queue(VOQ),the virtual switch queue(VSQ)is based on the destination switch to classify packets.The VOQ and VSQ are together added to the queue structure.If the network traffic model is unknown,the queue length of the virtual switch queue can be used to sense the size of the traffic between the switches.When the traffic size is basically unchanged,the routing algorithm guarantees that the packet does not go out of order when reaching the destination switch.The energy-saving module in the network obtains the information of the queues in the ports,compares the dynamic switching thresholds,and controls the exchange unit on and off in real time.The network performance is taking into account and the multiplexing ratio of the switch is improving.Switching algorithms also add a mechanism to prevent frequent switching.Simulation uses Bernoulli traffic and the burst traffic model to set different proportions for unicast and multicast mixed services.The results show that the algorithm closes more idle switches in the case of unicast and multicast mixed services,compared the network without energy saving algorithm.There are good energy-saving effects under test.Because the load balance in the network is closely related to the energy saving effect,this thesis proposes an energy-saving routing algorithm based on load balancing.This algorithm abandons the single goal of minimizing the energy consumption in a network,and incorporates load balancing and switch energy consumption in the optimization goal to choose an optimal path.The main contents of the part are:The relationship between load balancing and network energy saving is clarified,and a load balance calculation method is designed.With the link load balance calculation module and the optimized route calculation module in the energy saving system,an appropriate path is selected for the packet.To avoid selecting links with heavy load,the algorithm adds the queue length threshold to the route calculation module.In order to meet the connected state in the network,the opening and closing of the core switch also refers to the switch status of the aggregation switch and its ports,and the switch action between the upper and lower tiers has greater linkage.Simulations were performed by mimicking traffic patterns in the data centers.The results from a number of parameters showed that the algorithm has good results in balancing network energy conservation and load balancing.