Research On Rendezvous-based Caching And Forwarding In Named Data Networking

With rapid development of the Internet,there are more and more content-based applications,such as multimedia content sharing and media streaming applications.The Internet has become a distributed system that uses various tools and mobile terminals for information sharing.People are no longer concerned about where the content is,but what the content is.However,traditional TCP/IP network is designed to solve the addressing and communication issues,which exposes a series of problems when faced with content delivery.To address these problems,Information-Centric Networking was proposed.Named Data Networking is a well-known network architecture inheriting the idea of ICN,which has the natural advantage to support large-scale content delivery and mobility of terminals due to its name-based routing,receiver-driven transport pattern,inherent multicast support,in-network caching,adaptive and hop-by-hop forwarding.However,different from the fixed-length IP addresses,the content namespace of NDN is much larger than the IP address space.Names in NDN are hierarchical and are similar to URLs,which consist of several unfixed-length name components.These features bring great challenges to the storage and lookup of FIB,PIT,and CS in the NDN forwarding engine,making it difficult to guarantee the line-speed forwarding of packets.Therefore,it is particularly important to optimize the storage and lookup performance of the NDN forwarding engine.The main contributions of this thesis include the following four aspects:Firstly,as the content each NDN router is likely to handle are the whole content namespace,each FIB entry is likely to be used;if Data doesn't return as soon as possible,the size of PIT will get larger and larger;the size of CS is limited,low hit ratio may cause frequent cache replacement.To solve these problems,this thesis proposed a rendezvous-based caching and forwarding strategy that aggregates requests for same and similar contents to the same rendezvous point,so that each router deals with only a subset of the whole content namespace,which leads to the following benefits:1)the hit ratio of CS increases and the replacement of CS decreases;2)the size of PIT gets smaller;3)the number of used FIB routes in each NDN router gets smaller,making it possible to use route cache.Secondly,this thesis proposed a mechanism that enables tunneling in NDN to support the implementation of the rendezvous-based caching and forwarding strategy proposed in this paper,and the proposed mechanism is fully compatible with NDN's forwarding logic.The NDN tunneling mechanism also contributes to the implementation of services that depend on third parties,which improves the scalability of NDN and brings opportunity to new network functions.Thirdly,this paper proposed to apply route cache to NDN FIB design based on the high temporal locality and spatial locality of the Internet traffic.It loads the recently used FIB routes to route cache so that these routes may satisfy future requests.By studying IP route cache,this thesis identified the root cause of the cache hiding problem due to longest prefix matching,i.e.,the dependence between the cached FIB routes and the uncached FIB routes.Two approaches named atomic route caching and on-the-fly route caching were proposed to solve the problem.Finally,this thesis extended the NDN,a well-known ICN network architecture,and modified the ccnSim simulation framework to implement the rendezvous-based caching and forwarding strategy,as well as the two route cache solutions.Trace driven simulations using an HTTP request trace from China Mobile and an HTTP trace gathered by UC Berkeley demonstrated the feasibility and effectiveness of the design.