Design And Implementation Of Optical Interconnection Node For Distributed Computing System

With the development of mobile Internet and artificial intelligence,the introduction of large-scale new applications in 5G era has made the computing scale and network traffic of data center(DC)increased rapidly.With the development of distributed computing applications,the communication modes of traffic in data center are becoming more and more diverse.Distributed computing systems proposed new requirements for the network structure.Traditional networks are increasingly unable to support the new data-intensive applications.As a new type of optical interconnection network,disaggregated optical networks and resource disaggregated networks have attracted more and more attention.These new types of optical interconnection networks recommend that the existing network structures and functions of nodes become open.In view of the new challenges posed by open interconnection networks and resource disaggregated networks,this thesis proposed an optical interconnection node of disaggregated networks for distributed computing systems.We investigated open system functions for disaggregated network interconnected nodes,including data level functions such as flow information statistics,transmission channel selection,and optical packet aggregation,etc.Reporting of in-nodes flow information and states of each function module at the control plane.Moreover,configure the parameter in the node and the optical switch matrix connected to the node according to the instructions of the upper control system.We realized the function logic above to verify the function of the node through employing the high-speed FPGA to build interconnection network experiment platform.In addition,based on the optical interconnected node above,we proposed flow-aware fault diagnosis and recovery,analyzed the flow classification through the correlative feature of flows and extract the path information each flow takes,then realized the fault diagnosis and recovery through the SDN controller.We verified the function above under different types of fault scenarios by our experimental platform.Finally,the network model and resource structure of the resource disaggregated datacenter(DDC)network are analyzed.We discussed the relationship between memory traffic and network traffic of several distributed computing jobs under different DDC model structures.Then we simulated the traffic pattern in the resource disaggregated network nodes based on the FPGA platform.Furthermore,we discussed the traffic queuing and scheduling problem in resource disaggregated networks.The impact of three common scheduling methods,namely,priority mode,bandwidth reservation,and fifo,on the traffic transmission performance of resource disaggregated network traffic is evaluated.