【作者】 王焱；

【导师】 曾家智；

【作者基本信息】 电子科技大学 ， 计算机应用技术， 2008， 博士

【摘要】 TCP/IP随着互联网的迅速发展而广泛流行,取得了巨大的成功,成为事实上的工业标准,在网络界形成了一统天下的局面。但随着网络规模的急剧扩大和各种应用的开发,此时Internet的发展已经超出了最初体系结构的框架,体系结构上的先天不足和后来各种技术的无序发展,使TCP/IP的缺陷和弊端不断暴露出来,人们逐渐认识到网络体系结构的重要性。目前国内外对网络体系结构的研究表现在两个方面:一方面是对TCP/IP进行改进,但是难以解决TCP/IP先天存在的QoS、安全性和可扩展性等问题。另一方面是针对网络技术发展的特点,结合网络应用的需要,进行新型网络体系结构的研究,从而解决TCP/IP的根本问题。本文论述的服务元网络体系结构就是一种全新的非层次型网络体系结构。论文首先详细分析了TCP/IP网络体系结构存在的问题,并阐述了新型网络体系结构的发展状况,然后对服务元网络体系结构及其关键技术进行了深入研究。本文的主要创新点包括:1.首次提出了服务元网络体系结构的网络构建规则、设计方案和实现模型。在曾家智教授提出的服务元网络体系结构思想的基础上,本文提出,根据网络需求,通过服务组合将服务元组织在一起,从而实现服务元网络体系结构的网络构建。依据该构建规则,本文进而提出了服务元网络体系结构的网络设计方案和实现模型。2.提出了基于端口间虚电路的QoS机制。传统的虚电路是节点间虚电路,不能区别对待相同节点间不同端口的数据流,因此无法保证业务端到端的QoS。而在基于端口间虚电路的QoS机制中,套接字直接与端口间虚电路标识绑定,因此可以为相同节点间不同端口的业务提供不同的服务,从而保证网络业务的端到端QoS需求。3.本文分析了端口间虚电路的数据传输规律,在此基础上提出了两种可靠数据传输机制。(1)提出了针对非紧急文本数据的自适应累计确认的滑动窗口协议SASW-AA。由于端口间虚电路传输数据的路径固定,因此数据包的到达顺序与发送顺序一致,数据包不会发生错序。根据这种数据传输规律,本文提出可以用一个ACK自适应地累计确认一组数据包,从而解决传统滑动窗口协议确认帧频繁导致上行链路拥塞的问题,而同时又能保证数据的可靠传送。(2)提出了针对紧急文本数据的立即重传的滑动窗口协议SWIR。与数据报方式不同,在端口间虚电路传输方式中,只要接收方收到失序的数据包,就可以断定有数据包丢失了。针对传统的滑动窗口协议存在着不能及时重传数据包的问题,本文提出了立即重传丢失数据包的方案,避免了不必要的等待时间,降低了紧急数据的传输延迟。4.本文提出了三种新的生存性算法。(1)提出了混合共享路径保护算法HSPP。为了解决通路保护(Path Protection)方法备份路径建立成功率低的问题,HSPP算法首先使用通路保护方法建立备份路径,如果建立失败,则将工作路径动态地划分为工作路径段,在划分工作路径时,需要保证其备份路径段跳数不能超过限制值。HSPP算法不但提高了备份路径的建立成功率,而且能够在恢复时间与资源利用率之间获得更好的折中。(2)提出了一种新的带有共享风险链路组(SRLG)约束的路由选择算法NRSSC。与传统算法不同,NRSSC算法将SRLG作为链路代价公式的一部分,在选择工作路径时主动避免SRLG链路,较之传统算法,有效提高了成功保护率。而且由于考虑了区分可靠性(DiR)要求,在提供不同级别保护服务的同时,提高了网络资源利用率。(3)提出了快速重构恢复算法FRRA。FRRA算法也是在故障发生以后建立恢复路径,但与以往恢复(Restoration)算法不同,该算法不用等待路由表收敛,不需要网络中其它路由器的路由信息,而是由保护入口路由器根据本地信源树、故障信息和路由策略独立确定恢复路径,因此大大降低了恢复时间,减少了丢包率。本文提出的服务元网络体系结构的可靠数据传输机制和生存性算法也同样适用于其它一些网络。其中,可靠数据传输机制还适用于ATM、MPLS等以固定路径传输数据的网络,特别地,用于以多播树方式传输数据的多播网络,可以大大简化现有的可靠多播协议。而本文生存性算法还可以用于MPLS、GMPLS等网络。更多还原

【Abstract】 TCP/IP has been prevalent with the rapid development of Internet and achieved great success. It has become the actual industry standard and gotten dominant status in the network world. But with the fast enlargement of network scale and exploitation of diverse applications, now the development of Internet has overstepped the frame of original architecture. The innate shortages of architecture and out-of-order evolution of diversified techniques make the limitations and problems of TCP/IP network emerged continually. People have been cognizant of the importance of network architecture gradually. Now the researches of network architecture in and out of country are in two aspects: The one is to ameliorate TCP/IP. But this is hard to resolve the inborn problems such as QoS, security and expansibility of TCP/IP. The other is to investigate new types of network architecture to resolve the internal problems of TCP/IP according to the characteristic of network techniques and the requirements of network applications.Service Unit based Network Architecture discussed in this dissertation is a new type of non-layered network architecture. This dissertation first analyzes the problems of TCP/IP network architecture in detail and describes the development status of new types of network architecture. Then Service Unit based Network Architecture and its crucial techniques are investigated in deep.The major innovations and contributions of the dissertation are as below:1. The network construct regulation, design solutions and realization model of Service Unit based Network Architecture are proposed. Based on the idea of Service Unit based Network Architecture proposed by Prof. Zeng Jiazhi, this dissertation proposes that in order to construct network of Service Unit based Network Architecture, Service Units are organized by Service Combination according to network requests. Based on this construct regulation, this dissertation proposes the network design solutions and realization model of Service Unit based Network Architecture.2. QoS scheme based on Virtual Circuit between Ports is proposed. Traditional virtual circuit is between nodes and could not make a difference among data flows which are though the same nodes but belong to different ports. So it could not guarantee end-to-end QoS of applications. While in the QoS scheme based on Virtual Circuit between Ports, sockets are directly bound with the marks of Virtual Circuit between Ports. So different services could be provided to the applications that belong to different ports within the same node and end-to-end QoS requirements of application are guaranteed.3. Data transmission rule of Virtual Circuit between Ports is analyzed in this dissertation. Based on this analysis, two reliable data transmission schemes are proposed.(1) Self-adaptive sliding window with accumulative ACK (SASW-AA) protocol for non-instant text data is proposed. Because the path of Virtual Circuit between Ports is fixed, so the sequence of arrived packets is as the same as the sequence of sent packets, that is, the sequence of packets will not be disordered. According to this data transmission rule, in this dissertation, one ACK is used to accumulatively acknowledge a group of packets self-adaptively. SASW-AA protocol could not only resolve the problem of congestion at upper link due to frequent ACKs in traditional sliding window protocol but also guarantee reliable data transmission.(2) Sliding window with immediate retransmission (SWIR) protocol for instant text data is proposed. In the Virtual Circuit between Ports transmission mode, different with datagram mode, as long as disordered packets are received, it is concluded that some packets are lost. To resolve the problem that packets could not be retransmitted in time in traditional sliding window protocol, this dissertation proposes a solution of transmitting lost packets immediately. This solution could avoid unnecessary waiting time and reduce the data transmission delay of instant data.4. Three new survivability algorithms are proposed in this dissertation.(1) Hybrid Shared Path Protection (HSPP) algorithm is proposed. In order to resolve the problem that the success probability of establishing backup paths is low in Path Protection approach, in HSPP algorithm, the Path Protection approach is first used to establish a backup path for the primary path. If the establishment is unsuccessful, then the primary path is dynamically divided into segments. When dividing the primary path, the hop count of its backup segments should be not more than the limitation. HSPP algorithm could not only improve success probability of establishing backup path but also gain a better tradeoff between the recovery time and the resource utilization ratio. (2) A New Route Selection with Shared Risk Link Group (SRLG) Constraint (NRSSC) algorithm is proposed. Different with traditional algorithms, in NRSSC algorithm, SRLG is taken as one part of link cost function and the SRLG links are avoided when selecting the primary paths. Compared with traditional algorithms, the successful protection probability is efficiently improved. Because Differentiated Reliability (DiR) is considered in this algorithm, it could not only provide protection services of different levels but also improve the resource utilization ratio.(3) Fast Reestablished Restoration Algorithm (FRRA) is proposed. In FRRA algorithm, the recovery paths are also established after the failure has occurred. But it is different with previous Restoration algorithms in that it need not wait for the convergence of routing tables. Routing information of other routers in the network is not needed and the task of selecting recovery paths is implemented by Protection Ingress Router according to local source tree, failure information and routing policy. So the recovery time and the data lost ratio are greatly reduced.The reliable data transmission schemes and survivability algorithms of Service Unit based Network Architecture proposed in this dissertation are also suitable for some other networks. In them, the data transmission schemes are also suitable for networks that use fixed paths to transmit data such as ATM and MPLS network. Especially, when used in multicast networks that use multicast tree to transmit data, the reliable multicast protocol could be greatly simplified. The survivability algorithms proposed in this dissertation are also suitable for MPLS and GMPLS networks.更多还原