The Internet is regarded as the basic communication infrastructure for the new generation of information technology,e.g.,Internet of Things,Industrial Internet,big data,etc.The delay feature of Internet characterizes the information dissemina-tion in Internet and is a key performance index to evaluate the Internet.Recently,with the rapid development of new generation of computing technology,e.g.,high-performance cluster computing,cloud computing,edge computing/fog computing,etc.,the timeliness issue of the computing and development towards Internet ser-vice has been gradually transformed to the real-time data transfer issue in Internet,i.e.,the delay issue.Developed up to now,the Internet has gradually shifted from the original'space grab' mode to the 'time saving' mode.Research on infrastruc-ture and communication principles of the Internet towards real-time data transfer in Internet is of great significance for improving the efficiency of network resource utilization and service scheduling,and guaranteeing the real-time deployment of In-ternet service.This dissertation focuses on the delay issue in Internet and conducts the research on guaranteeing mechanism enabling real-time data transfer.With this researching target,this dissertation is carried out from the following aspects:delay detection,delay management and delay control.Firstly,an in-depth study about the delay detection issue in Internet is con-ducted,i.e.,characterizing the current delay in Internet.Based on the large-scale IPv4-level sample authorized by CAIDA,more than 9 million efficient detecting paths are extracted,and the analysis of the features of end-to-end delay in Internet and the evolutionary trends in time domain under Internet macroscopic topology is performed.It is found that the weak correlation between the end-to-end delay and the number of router the detecting packet traverses(i.e.,the hops),and the end-to-end delay is dominated by the delay of one or several links along the efficient detecting path.It is also remarkable that the bottleneck-delay defined in this dis-sertation appears in 70%of the efficient detecting paths,which is considered to have a strong impact on end-to-end delay.To further study the features of bottleneck-delay,an IP united mapping scheme which combines the IP centralized mapping with the IP geographic mapping is proposed.Statistical results show that two ends of a bottleneck-link are usually located in the same country(with probability 80%),and the links with a large number of entrances(in-degrees)and few number of exits(out-degrees)or the average shallower links are prone to leading to bottleneck-delays.At last,a deep discussion about the factor of bottleneck-delay is given,which demonstrates queuing delay is the key factor of bottleneck-delay mapped in the same country and propagation delay is the major component of bottleneck-delay for the long-range link.Afterwards,a new traffic scheduling engine DTE-SDN towards delay-sensitive traffic in Internet is proposed.DTE-SDN is dedicated to lead-ing to a delay management platform(including network resource manage-ment,equipment integration,traffic engineering(TE))that can guaran-tee the real-time data transfer in Internet.DTE-SDN utilizes the fea-tures of Software-Defined Networking technique with OpenFlow protocol(e.g.,Port_Status-Request?Port_Status_Reply?Packet_In?Packet_Out)to capture an overall view of the network and provides the ability of actively QoS metrics(e.g.,network link-delay,throughput)monitoring in real-time.To efficiently schedule the delay-sensitive traffic,a dynamic TE scheme capable of multi-routing is proposed to be deployed on the scheduling module of DTE-SDN,which allows DTE-SDN to compute the optimal TE strategy based on the instantaneous QoS metrics.Particu-larly,the TE issues can be addressed by addressing a proposed maximum flow over time issue based on time-expanded network,which can lead to a pseudo-polynomial algorithm.Furthermore,to distribute the traffic(especially for the UDP flow)a-mong multiple end-to-end paths and deploy TE,an improved group table module with probabilistic-matching technique of OpenFlow is proposed to be deployed in networking forwarding elements.Experimental results demonstrate that the pro-posed DTE-SDN is able to measure the throughput and delay under acceptable error range and can dramatically enhance the transfer efficiency than the tradition-al scheduling(e.g.,the scheduling based on shortest-path routing approaches or the maximum flow issues).Finally,this dissertation takes a deep discussion about the Multi-service Delay-constrained Transfer Scheduling(MDTS)issue.MDTS considers a delay control scenario where multiple flows related to different kinds of Internet service concur-rently require for real-time data transfer scheduling,i.e.,all the data transfers are confined to be completed within a time constraint.Based on the complexity proof for the MDTS issue,two static scheduling approaches(MDTS-R,MDTS-P)and a dynamic scheduling approach(MDTS-H)are proposed respectively,both of which are based on multi-path routing and flows over time technique.Particularly,MDTS-R utilizes maximum flow over time theory to address the flow optimization issue and is based on the single-service heuristic,while MDTS-P is derived from solving a pro-posed maximum constrained flow over time problem and optimizes the multi-service flow from the overall.MDTS-H makes use of maximum multi-commodity flow over time theory to greedily optimize the network flow on the whole,until a service's transfer is completed,leading to a new rescheduling.Especially in MDTS-H,the auxiliary-matrix is adopted for addressing the residual-flow issue during reschedul-ing,such that the minimum switching time(the time to reschedule)can be accu-rately computed.The simulational results demonstrate that,both of the proposed algorithms can address the MDTS issue in Internet.Especially,MDTS-R runs the fastest,MDTS-P is more efficient when the number of service is little.MDTS-H benefits from the dynamic rescheduling scheme and is more dedicated to addressing the MDTS issue with large number of service. |