The Research And Implementation On High-performance Name Lookup Technology Based On GPU

Named Data Networking(NDN)is a new network architecture that faced to future Internet,which is aimed at subverting traditional TCP/IP network whose communication model is based on address,then replace it as named data,thus fundamentally solve the problem of traditional network architecture in scalability,dynamic and security issues.In the routing and forwarding process of NDN,name lookup is a core operation,and it is also the key factor that affects the performance of NDN packet forwarding.However,because NDN names are complex in structure,indefinite in length,and have no theoretical upper bound,which makes storage compression and lookup acceleration become a key technical challenge in solving name lookup problems.Most existing name lookup solutions are optimized for only one aspect of storage compression or lookup acceleration,and they can't achieve good performance in both aspect.But Multi-Aligned Transition Array(MATA)achieve both of them with the help of the classic sparse table compression strategy and the powerful parallel computing capability of Graphic Processing Unit(GPU).However,the storage and utilization rate of this solution is still low.Therefore,based on a full understanding of NDN name lookup's requirements and GPU platform features,this paper designs a new type of data structure whose name is Candidate-Aligned Transition Array(CATA).In CATA,multiple candidate mechanisms are introduced to cope with multiple hash functions to improve storage utilization and then compress storage.Besides,CATA also draws support from GPU to accelerate lookup.The results of comparative experiments show that the CATA can achieve the same search speed compared to CATA,and at the same time,storage utilization of CATA can be increased to about 90%,and the total storage of cost less than a fifth of MATA.Furthermore,considering the validation of the name lookup algorithm is mainly based on simulation,this paper also focuses on the special structure of NDN packets and the general processing flow of CPU-GPU cooperative acceleration,and then implements an name lookup engine for NDN data packets.On this basis,the system level testing of name lookup algorithm is further implemented.Finally,to validate the actual performance of the name lookup engine in this paper,we also deploy a comparison experiment between this engine and another G-opt framework with name lookup system implementation.The experimental results show that the overall performance of the name lookup engine is better than that of G-opt,and the throughput is improved by 40%,while the delay is decreased by 15%.