Key Technologies Of Programming Model And Consensus Protocol In Named Data Networking

Named Data Networking(NDN),as an Information-Centric Networking architecture,uses data names instead of IP addresses for flowing data.This conceptual change improves content distribution efficiency,network security and mobility compared to TCP/IP,and also brings many challenges.NDN introduces new design patterns for applications,but having no transport layer protocol support increases the difficulty of application devel-opment.What's more,the absence of consensus protocol leads to the lack of reliability guarantee for large-scale distributed systems in NDN.To address these problems,this the-sis presents a thorough study of NDN programming model,video streaming applications,consensus protocol and replicated state machine.Main contributions of this thesis are summarized as followings:(1)To address the programming difficulty problem,this thesis proposes a program-ming model based on Application Data Unit(ADU)exchanging,to ensure the data transmission reliability and development efficiency.For largely asynchronous publishing,it provides throughput perceptive parallel ADU consumption;for real-time publishing,it provides an adaptive ADU pipeline strategy.Experimental results show that the parallel strategy helps retrieve the data with the required throughput by adjusting the parallel window.Compared to newly generation of data packets for each request,the producer sender buffer can reduce up to 95%computational overhead.The Manifest technique further enhances the efficiency of ADU publishing.(2)To verify the functionality and performance of the programming model,this thesis designs and implements live(NDNlive)and pre-recorded(NDNtube)video streaming applications over NDN.Both of them publish and retrieve video data in the unit of frames,which is the true information-centric fashion by corroborating the ADU concept of NDN.NDNlive and NDNtube verify the ADU pipeline and parallel retrieval strategies,respectively.They have been deployed over the official NDN Testbed.Experimental results show that they can provide fluent and synchronized video streaming across 11 time zones in the worldwide while tolerating small network problems.(3)To provide the reliability guarantee for distributed systems in NDN,this thesis proposes Naxos,a consensus protocol for NDN using Paxos theory,which explores the features of Pull-based network model and NDN naming conventions to improve latency and throughput.With reference to the established cost model,Naxos adopts a novel Self-learning process to reduce the workload of the distinguished proposer and balance the load between multiple nodes.Experimental results show that Naxos can achieve up to 50%?69%commit latency improvement in wide area and up to 4.7?5.57 times throughput improvement in a cluster.(4)To provide fault-tolerant service,a replicated state machine based on NDN(NaxosRSM)is proposed.For applications that only require eventual consistency,the system provides a Quick Read strategy to further improve the throughput.NaxosRSM adopts proper NDN naming conventions and forwarder configuration to make group changes completely transparent to clients to achieve better availability.Experimental results show that Quick Read maximizes the throughput increase in proportion to the number of nodes which serve read requests.In the case of node failure,NaxosRSM achieves lower performance degradation than Multi-Paxos RSM for TCP/IP,which reduces up to 80%out-of-service duration.