Research On Reconfigurable And Virtualizable Router Arichitecture And Key Technology

With the development of networking technologies,the number of network business has shown explosive growth.Diversified businesses require network operators to provide a vairty of service capabilities.However,existing network operators' infrastructure is suffering from the ossifications caused by closed and dedicated soft-/hardware based network appliance.These ossifications have hindered the development of network technology.Network infrasturture with elasticity,flexibility and definability is considered as the future direction.Reconfigurable virtual router could provide reconfiguability and virtulizaiton capabilities,and support multiple virtual instances on a shared platform.Therefore,the reconfigurable virtual router is known as the effective approach to advance the network's elasticity,flexibility and definability.In this regard,domestic and foreign research groups for reconfigurable virtualized router architecture and key technologies put forward a lot of solutions.However,these schemes have many shortcomings,which are as follows: 1)The limitation of the traditional router architecture limits;2)the lack of data plane functional path for heterogeneous network functions;3)no method to allocate physical resources to the actual demand of heterogeneous network functions;4)In data plane packetisolation scheduling,the lack of support for heterogeneous network traffic isolation between the multiple virtual data planes.In view of the above-mentioned problems,this paper starts from the reconfigurable virtualized router architecture and researches some key technologies around the data plane.The main research works are as follows:1)Propose a novel kind of reconfigurable router architecture,called Router is Data Center,RiDC for short.RiDC redesigns the traditional router architecture by reference to the idea of Cloud Computing and Data Center,separating the router structure into a variety of standard components,including routing control components,packet processing components,linecard switching component and intraconnecting components.Through defining the standardized interoperability interface between different components,RiDC breaks the venders' monopolies for the network equipment and achieves the building block implementation of network equipment.2)Existing solutions could not provide the effective approaches for data plane function topology reconfiguration.To this end,we propose a DAG-based(Directed Acyclic Graph)to model the network function topology.Based on the model,we propose the function reconfiguration algorithms and provide the corresponding validity of the algorithms.At last,we designed the function reconfiguration framework on a multi-core server.The experiment results show that our framework is able to support function reconfiguration,in the context of keeping the throughput of deployed network functions.3)Current computing resource allocation algorithms could not provide the effective approaches for data plane preference reconfiguration.To this end,we propose the computing resource allocation algorithm among multiple functions.The allocation weight depends on the requirements of computing resources based on different functions' CPU cost caused by the different traffic attributes.The goal of our allocation algorithm is to balance the workload between two directly connected functions.Compared with existing algorithms,our algorithm is able to ensure the throughput requirements of each function,on the premise of maintaining higher physical resource utilization.4)Existing packet scheduling algorithms could not satisfy the isolation requirements of RiDC data plane.To this end,we propose a novel packet scheduling approach.The idea of our approach is to periodically distribute resources tokens among multiple virtual data plane.If a virtual plane consumes all its tokens,all its packets would be dropped.According to the different requirements of ingress and egress isolation,we propose two isolation scheduling algorithms.The ingress scheduling algorithm is based on multi-resource constraints and egress isolation is only based on bandwidth constraints.Through consolidating the two algorithms,RiDC is able to ensure isolation in both ingress and egress.Finally,supported by the national 973 project “Research on reconfigurable information communication basal network architecture” and its topic “Testbed for reconfigurable information communication basal network”,we realize a RiDC prototype.The prototype is built with the COTS hardware commonly used in current Data Center,such as multi-core servers,10G/1G Ethernet switches or popular OpenFlow switches.Through defining the open inter-operation protocol among the physical components,our prototype achieves high modularity.Through integrating existing IT virtualization technologies,we implement the virtualization environment.Evolution results show that RiDC prototype owns line-speed packet processing rate and isolation capability.