Research On Network Structure And Control Plane Techniques For Optical Interconnected Data Center Networks

Cloud computing is a computing model that users can conveniently access a shared resource pool on demand through the Internet at anywhere and on anytime.Data center network(DCN)is a network that interconnects a large number of servers in a data center and is used for the transmission and exchange of a large amount of data.With the development of cloud computing,data center network has become a hot topic.Traditional data center network is a hierarchical tree structure based on electrical Ethernet switches,but it suffers the problems such as low bisection bandwidth,high power consumption and large packet delay.Optical network has the advantages of high bandwidth and low latency.Therefore,hybrid electrical/optical DCN and all-optical DCN structures have been proposed.In these structures,the optical network is mainly applied to the inter-rack communication,and the intra-rack communication still adopts electrical switches at the top of rack(ToR switches),which causes many problems.The energy consumption of ToR switches accounts for 90% of the total energy consumption of network equipment in DCN.The traffic within the rack accounts for 80% of the total network traffic of DCN.Therefore,it is necessary to design the optical interconnection structure and corresponding control plane mechanism within a rack for DCN.This thesis first designs a coupler-based software defined passive optical intra-rack network(SD-POIRN)structure.The data plane adopts couplers to interconnect servers in the same rack and provides four server light source plans.The control plane uses a software-defined centralized controller to coordinate intra-rack communication.In order to avoid intra-rack traffic collisions,a software defined medium access control mechanism SD-MAC is proposed.SD-MAC utilizes the server grouping and the Max-Min fair share bandwidth allocation algorithm to allocate time and wavelength resources fairly among servers.Simulation results show that when the number of servers in rack is 64,the energy consumption of SD-POIRN is 78.4% lower than that of traditional ToR switch solution.SD-POIRN uses 2×2 couplers is more scalable compared with 3×3 couplers and a rack can support 128 servers.SD-POIRN can improve network throughput and packet delay performance.In addition,this thesis proposes a wavelength selective switch based intra-rack network WSSIRN structure.The data plane uses the cascaded fabric of the wavelength selective switch and the coupler to interconnect servers in the same rack.WSSIRN can not only provide unicast,multicast and broadcast communication between servers in the same rack,but also provide servers inter-rack communication.The control plane uses a software-defined wavelength scheduling mechanism.It uses round robin scheduling and Max-Min fair share bandwidth allocation algorithm to allocate wavelengths to servers.Simulation results show that when the number of servers in a rack is 64,the cost and power consumption of WSSIRN are reduced by 25.9% and 74.1% compared with the traditional ToR switch solution.WSSIRN can achieve higher network throughput than the SD-POIRN structure.The simulation results show that the optical interconnection data center network structure and control plane mechanism designed in this thesis has advantages in cost,energy consumption and network performance,which lays a foundation for the future research and development of data center network.