Research On NDN Performance Optimization Technologies Based On Programmable Data Plane

With the emergence and application of technologies such as Internet of Things,Internet of Vehicles,Named Data Networking(NDN)has become a research hotspot in emerging network architectures owing to inherently in-network caching,efficient data distribution and other vital technologies.However,TCP/IP protocol is rooted in existing network equipment,which makes the practical deployment of NDN have compatibility problems in the actual network environment.The emergence of programmable switching chips and high-level data plane programming languages enables network administrators to customize protocols and forwarding processing logic.It can improve the data plane’s programmability and packet processing flexibility,providing an effective solution to the challenges faced by large-scale NDN deployments.Nevertheless,there are still numerous challenges to deploying NDN applications in the programmable data plane,mainly in the following three aspects.(1)Routing efficiency.The existing NDN routing mechanism can only search for on-path node caches and cannot fully utilize off-path cache resources;(2)In-network cache effectiveness.The traditional caching strategy still suffers from low cache utilization,which can easily cause network redundancy and limit the in-network caching effectiveness of NDN;(3)Data distribution performance.The high mobility and real-time characteristics of NDN require frequent interaction between the control plane and the data plane.The controller issues enormous control flow tables,resulting in increased processing latency.This thesis focuses on the NDN performance optimization technologies based on programmable data plane,the research contents and main innovative work of this thesis are as follows:(1)Content transmission path based multi-path routing schemeThis thesis proposes a multi-path routing scheme based on content transmission path to address the problem of low cache utilization due to inefficient use of in-network cache resources by routing mechanisms.By recording the out/in ports of the previous content at the node,when a content request with the same name arrives,it is routed first to the recorded port to query more nearby nodes that may have cached the content copy.This mechanism can effectively reduce the load on the content source server and improve the cache hit rate.Experimental results demonstrated that using a multi-path routing mechanism based on content transmission path can reduces the average routing hops by about 25% and increases the in-network cache hit rate by11%-23%.(2)Cache-optimization strategy based on dynamic popularity and replacement valueThis thesis proposes a cache optimization strategy based on dynamic popularity and replacement value to solve the problem that limited cache capacity limits NDN performance.The node calculates the content popularity by considering the number of requests for the content.It dynamically adjusts the cache threshold to store the content with the highest probability of future requests.When cache replacement is needed,the replacement value of the content is calculated by considering the last request time,popularity and transmission cost,and the content with the minimum replacement value will be moved out of the cache.The experimental results show that compared with the traditional caching strategy,the cache hit rate of the cache optimization strategy proposed in this thesis improves by 11.6% on average,and the average number of route hops is reduced by about 46%,which effectively improves the caching performance.(3)Protocol-independent bitwise multicast distribution technologyThis thesis proposes a scalable bitwise multicast technology in NDN to resolve the problem of increased data transmission latency resulting from the controller issuing enormous flow tables for data distribution.The bitwise multicast distribution technology considers the characteristics of NDN native support for multicast,defining a multicast port mask field in the P4 program of the programmable data plane to identify the destination port to which the packet will be forwarded when it returns.The mechanism redefines the pending Interest table to maintain the same bit vector as the port mask,enables data distribution through software methods when packets are returned,and implements the forwarding processing logic for two different types of packets in a programmable device.Experimental results indicated that compared to dynamic multicast,bitwise multicast technology significantly reduced the number of flow tables and reduced the average response latency by about 57% compared to unicast methods.