[ English | Japanese ]
Back to Index Page

ACADEMIC YEAR	2004
NAME	IMAIZUMI, Hideaki
TITLE	A study of Internet End-End Path Control Architecture
ABSTRACT	In this dissertation, a path control architecture which enables users to establish preferable end-end path over the Internet satisfying user's demand on quality of service and type of the path is discussed. Internet has become an indispensable communication infrastructure allowing people all over the world to go about their social lives, regardless of location or time. Applications over the Internet, in general, are characterized by its required bandwidth and permissible delay. For example, when transferring a file, the available bandwidth is emphasized over delay, while for Voice over IP, delay is of more concern. For PC clustering over a network, a more strict condition is applied at both ends. The advancement of optical technology, especially Wavelength Division Multiplexing (WDM) technology, has drastically increased the available bandwidth of each fiber. In an experiment conducted by NEC in the year 2001, a single fiber was shown to provide a bandwidth up to 10.9Tbps. Lucent Bell laboratory estimates a single fiber is able to provide up to 100Tbps of bandwidth in theory. The available bandwidth can also be increased through use of multiple fibers. Meanwhile, transmission delay can be divided into physical transmission delay and packet processing delay. Physical transmission delay is the time necessary to transmit packets over the physical medium, and packet processing delay is the time necessary for a router to process the packet for transmission. In general, a large portion of the transmission delay is from the physical transmission delay. The minimum physical transmission delay can not exceed the speed of light, and therefore the direct distance from one end to another is the route with absolute minimum delay, and this delay increases as the route becomes longer. Therefore, for applications which require small permissible delay such as Voice over IP, data traffic must be routed through the path with the shortest physical distance. However, a hop-by-hop packet switch networks such as IP do not handle explicit routes based on applications flows. The route for each packet is based on the source and destination of a flow, and the different characteristics and requirements of applications are not considered. Therefore, even if a route satisfying permissible delay for an application exists, packets are not always transmitted through such route. Multi-Protocol Label Switching (MPLS) is a method to explicitly use specific routes by identifying the flow in various precision. MPLS is based on circuit switch network technology and uses IP networks to establish a circuit through IP networks. MPLS has been generalized to GMPLS for adapting to Fiber Cross Connect(FXC), Optical Cross Connect(OXC) and TDM slots. GMPLS is expected to establish a dynamic optical path through intra- and inter-carrier. Several studies are working to establish end-end path on the Internet using GMPLS. Some focus on establishing the end-end path using not only GMPLS but also MPLS. However, these studies have two major problems: (1) The studies are based on MPLS which is known for its high cost in terms of deployment. MPLS is not applied to all networks and therefore users in non-MPLS networks can not use end-end path provided by these studies. (2) These studies focus on stable bandwidth allocation and do not consider transmission delay. Therefore an application which is sensitive to delay can not utilize the path with the minimum delay. In this dissertation, a novel technique, called Flow-Mapped Explicit Host Routing (FMEHR), which transmits packets on an explicit route by using an unused address space is proposed to solve the former problem. This technique realizes a function similar to MPLS without changing the forwarding paradigm. Therefore, the cost of applying MPLS to a network is decreased compared to MPLS. Using FMEHR enables usage of different multiple links, which makes it possible to configure a path consisting of different links, in cases such as durable and non-durable links, to pass a router. A flow-map method is necessary for realizing FMEHR in an IP network, and the following four are given: Egress-router storing method, trailer storing method, source route method, and tunnel method. As a result of comparing the characteristics of the above methods with MPLS, trailer storing method had an advantage in many ways, although in cases where overhead caused by IPv6 routing header can be avoided inside the network when a jumbo frame of Ethernet is available, source route method was more preferable in terms of migration. To solve the latter problem, path control architecture for the Internet consisting of ASes with MPLS or FMEHR applied is proposed, and components to realize the architecture are discussed. As one of the most important components, graph information of a network inside an AS with attributes such as link delay are required to manage the network. However, there are some problems to obtain the information on a FMEHR network; this dissertation designs a link-state routing framework to solve these problems, and the approaches to solve them are described below. A FMEHR network is independent from data link technologies and therefore consists of various types of links including satellite links, account links and multiple links between same two routers. Current link-state routing protocols are not designed for application to such networks and therefore can not avoid consumption of unnecessary bandwidth caused by control traffic. On the other hand, in the current link-state routing technologies, various types of link-state routing protocols exist. Since each protocol has different characteristics, each uses different identifiers, IP or NSAP addresses, or user given identifiers to express graph information. When an appropriate link-state routing protocol for each sub-network consisting of an AS is applied, obtaining the whole graph information in the AS is difficult because each sub-graph information is expressed by different identifiers. To solve these problems, the element that determines the characteristics of a link-state routing is decided and a framework is designed in this dissertation. The framework does not depend on a particular network layer protocol nor depend on particular topology broadcast protocol to obtain graph information inside a link-state routing protocol. On this basis, each problems are solved by developing the components for the here mentioned framework. By focusing on topology broadcast, which makes up a majority of control traffic created by link state routing, we designed in detail a topology broadcast protocol called Graph Configurable Topology Broadcast Protocol(GCTBP) GCTBP, which prevents the flooding of control traffic to multiple or specific links. This dissertation approaches such problem by using GCTBP, which is an extended protocol of Topology Broadcast Reverse Path Forwarding(TBRPF). Simulations on random topologies with two-level priority showed that the technique was effective on link-attribute change event in terms of communication cost and convergence time. Moreover, to solve the problem of graph information compatibility, a new method of protocol development, enabling virtual networks to easily adapt to other network technology and heterogeneous network environments is proposed. This dissertation describes (1) ways to emulate a virtual datalink ANCIF, which uses identifiers independent of particular network technology in variable real networks (2) protocol development using ANCIF. Implementations of ANCIF applied to real networks and the evaluations of overhead resulting from virtualization of ANCIF are also shown. The result is small enough not to affect performance. Lastly, this dissertation discusses a middleware system RBDBS, which enables applications to choose an appropriate topology broadcast protocol based on the class of the target network, and APIs for the applications that are realized. The overhead resulting from the middleware approach is evaluated and the result is small enough not to affect performance. As a conclusion, this dissertation contributes to realize an Internet path control architecture, enabling a user to freely establish a preferable end-end path. The next-generation Internet becomes a hybrid infrastructure of packet switch and circuit switch networks. This dissertation proposes a novel technique, called FMEHR, to construct path on a packet switch network without MPLS. Moreover, path control architecture for the Internet consisting of ASes with MPLS or FMEHR applied is proposed, and the components to realize the architecture are discussed. This architecture is realized by the advanced link-state routing framework to manage the whole graph information inside an AS.
CONTACT	To obtain the whole paper (in Japanese), please contact; IMAIZUMI, Hideaki (hiddy@sfc.wide.ad.jp)

Back to Project Home Page