Research Of Flow Scheduling In Green Data Center Networks

Lots of approaches about saving energy of data center(DC)have been proposed to reduce the operating cost of cloud computing in recent years.Those methods not only include architecture of DC,but also incorporate resource allocation,virtual machine placement and flow scheduling.Though those studies could decrease the consumption of DC,most of methods only take energy-saving or improving the quality of service(QoS)into account,which leads to that it can't balance the consumption and QoS of DC.Thus how to balance the energy consumption and QoS of DC becomes the new hotspot.The paper mainly analyzes the energy consumption of the typical DC networks,and proposes the approaches of reducing the consumption and improving the power usage effectiveness(PUE)of DC from the perspective of network routing and flow scheduling.Firstly,it analyses the characteristics of the DCs and some energy-saving methods,and points out the advantages and disadvantages of those approaches,then proposes a flow scheduling algorithm:flow consolidation and flow preemption(FCAFP),to overcome the shortcomings of the above approaches.The characteristic of the algorithm is that it could reduce the consumption and insure the QoS of DC at the same time.The simulation results show that FCAFP can improve PUE and link utilization of DC networks,and reduce the consumption significantly while just has little impact for the QoS of DC.Secondly,it proposes a dynamic flow scheduling algorithm based on game theory:DFSGT.It analyzes the characteristics of the data stream,and takes different requirements into account.Then it divides the flows of DC into different levels to transmit,and considers the demands of operators and users.Therefore,it not only meets the users' requirements:low cost,high performance,but also increases the revenue for operators,and reduces the consumption of DC at the same time.DFSGT could find an appropriate equilibrium to balance the energy consumption and QoS to meet the requirements of all parties,which is the best characteristic of the algorithm,and it also could improve the resource utilization and the PUE of DC.