【正文】 performance issues. Good models of the topological structure of a work are essential for developing and analyzing interworking technology. This article discusses how graphbased models can be used to represent the topology of large works, particularly aspects of locality and hierarchy present in the Inter. Two implementations that generate works whose topology resembles that of typical interworks are described, together with publicly available source code. 1 Introduction The explosive growth of working, and particularly of the Inter, has been acpanied by a wide range of interworking problems related to routing, resource reservation, and administration. The study of algorithms and policies to address such problems often involves simulation or analysis using an abstraction or model of the actual work structure. The reason for this is clear: works that are large enough to be interesting are also expensive and difficult to control。 the typical host will be represented as a leaf connected to a single router node. Additional information about the work can be added to the topological structure by associating information with the nodes and edges. For example, nodes might be assigned numbers representing buffer capacity. An edge might have values of various types, including costs, such as the propagation delay on the link, and constraints, such as the bandwidth capacity of the link. The purpose of this article is to review the basic topological structure of the Inter, then present a modeling method designed to produce graphs that reflect the locality and hierarchy present in the Inter. Two implementations of the method are available。s Inter can be viewed as a collection of interconnected routing domains. Each routing domain is a group of nodes (routers, switches and hosts), under a single (technical) administration, that share routing information and policy. Each routing domain in the Inter can be classified as either a stub domain or a transit domain. A stub domain carries only traffic that originates or terminates in the domain. Transit domains do not have this restriction. The purpose of transit domains is to interconnect stub domains efficiently。 singlehomed stubs connect to only one transit domain. Some stubs domains may have links to other stubs. Transit domains may themselves be anized in hierarchies, . MANs connect mainly to stubs domains and WANs. Existing Topology Models One of the most monly used models for generating random works algorithmically is due to Waxman [3]. The nodes in the work are placed at random points in a two dimensional grid. Links (represented by edges between nodes) are added to the work by considering all possible pairs of nodes and then deciding whether a link should exist according to a probability function involving how far apart the two nodes are and how many links are expected to be in the whole work. The original intention of this approach was to generate works for paring Minimum Steiner Tree algorithms. It has several serious drawbacks when used for generating typical 10 inters. First, the works don39。s method have been proposed by the authors and others. Some of these attempts to restrict the longest links in the work, while others reduce the number of edges from any particular node. Still other modifications introduce a simple hierarchy to the work. None of them produce convincingly realistic works. 2 A Better Method Over the past few years a better method has been devised independently by the authors for generating graphs that reflect the hierarchical domain structure and locality that is present in the Inter [1, 4]. Three levels of hierarchy are modeled, corresponding to transit domains, stub domains, and LANs attached to stub nodes. The method constructs the graph piecewise, where the pieces correspond to domains at the different levels in the hierarchy. The connectivity within a domain (intra work connectivity) is dealt with separately from that between domains (interwork connectivity). Parameters Two sets of parameters control the coarse properties of the works generated. These parameters are chosen to provide reasonably simple control over the important structural characteristics of the graph. The parameters chosen also have obvious effects on the works that are produced. The first set of parameters governs the relative sizes of the three levels in the hierarchy: T, the total number of transit domains, and NT, the average number of nodes per transit domain. Note that= 1 and NT=1. S, the average number of stub domains per transit domain, and NS, the average number of nodes per stub domain. Note that S= 1 and NS=1. L, the average number of LANs per stub node, and NL, the average number of hosts per LAN. LANs are modeled as star topologies with a router node at the center of the star and the host nodes each connected to the center router. As pared to using a plete graph connecting all hosts in the LAN, this significantly reduces the number of edges in the graph and reflects the lack of physical redundancy in most LANs. Note that L= 0 and NL=1. The total number of routing nodes, NR, and the total number of hosts, NH are given by: NR =TNT(1 + SNs) NH =TNTSNSLNL Note that the parameter values are taken as the basis for distributions used to obtain the actual value for each run of the algorithm. Extra information can be associated with each parameter to describe the distribution of the parameter. For instance, an upper and lower bound on the number of nodes in a stub domain and the function for distributing the value between the bounds could be described. 11 The second set of parameters governs the connectivity within a domain (intrawork connectivity) and the connectivity between domains at the same or higher and lo