Performance Optimization For Software-defined Optical Interconnection Data Centers

In order to satisfy the performance requirements of software-defined optical interconnect data center(such as scalability,low cost and low latency),this paper optimizes the data center network by designing distributed control layer architecture,aggregating flow tables of switches in data layer,and proposing efficient switching scheduling algorithm.The research achievements are summaried as following:1)In the control layer,we proposed an algorithm with distributed placement of controllers.The centralized placement of the controllers is distributed into the optical interconnect data center.The controllers transmit messages through the optical links in the backbone network to alleviate the pressure on the controller to process the multi-flow requests.It provides a path to the nearest controller for these flows,but it also increases the transmission delay of messages between controllers.In this paper,an ILP(Integer Linear Program)formulation and a heuristic algorithm are proposed to optimize the distributed placement of the controller,with the consideration of the cost of link delay,controller cost,etc.,and through compromising the transmission delay of synchronous messages among controllers and the transmission delay of requested flows.Experiments show that proposed algorithms minimize the total overhead of the system while satisfying the flow request in the data layer,which improves the scalability of the control layer.2)In the data layer,we proposed a flow aggregation algorithm.In order to control the scales of flow tables,a flow aggregation algorithm based on dynamic routing generation strategy is proposed.Under the factors of traffic characteristics,transmission delay,and network topology,the flow table compression and route optimization are jointly designed.Firstly,an ILP formulation is proposed to minimize the total cost by considering the tradeoff of the hardware cost of the flow tables and the average transmission delay cost of traffics,and the optimal route is generated for flow aggregation.In order to improve the efficiency of algorithm execution,the Overlapping Routing with Weighted Flow Aggregation(ORWFA)is proposed.The core idea is to build a minimum Spanning Tree based on the principle of High Weight Flow First(HWFF).And then the generated routing based on minimum Spanning Tree is used to aggregate flows.The experiment results show that proposed algorithms improve the aggregation rate of the flow tables and reduce the memory chip cost in flow table through the dynamic routing allocation strategy.3)In the data center,we proposed a multi-hop scheduling algorithm for data packets.High-performance data centers require high-capacity optical switches to guarantee Quality-of-Service(Qo S).In multilevel switching architecture,the reconfiguration cost of optical switches and the efficiency of scheduling algorithm are two key challenges.This paper utilizes multiple intermediate-level optical switches operating in parallel to provide a spatial speedup to compensate for reconfiguration overhead.In addition,a multi-hop scheduling mechanism is proposed,which uses the empty time slots in the matrix decomposition scheduling algorithm based on non-blocking guarantee.The MHRS(Multi-Hop Routing and Scheduling)algorithm and DETOUR algorithm are designed respectively.Simulation results show that both algorithms can effectively reduce the average transmission delay of packets.