ITR: Collaborative Research in Internet Topology Models - A Foundation for Large-Scale Simulations

对于试图了解网络体系结构和协议设计原理的研究人员来说，网络模拟是不可或缺的工具。在任何中等规模到大规模的模拟中，一个关键参数是拓扑，即网络节点的组织和相互连接的方式。良好的拓扑学模型对于良好的模拟是必不可少的。PI开发了图形建模软件，该软件目前被广泛用作生成拓扑的工具，特别是大型网际网络的模型。佐治亚理工学院互联网拓扑模型(GT-ITM)包允许研究人员构建其结构类似于互联网节点级结构的模型拓扑：路由器或交换机，通过(双向)链路连接，并分组到域中。GT-ITM软件包含在“NS2”[2]中，“NS2”[2]是事实上的网络模拟开源标准。尽管GT-ITM被广泛使用，尤其是它的中转末梢模型，但关于网络拓扑建模的一些关键和基本问题仍然没有得到回答。例如，1)拓扑模型。最近的数据表明，当前的Internet拓扑具有一些属性，这些属性在GT-ITM的传输存根模型中没有得到很好的反映[17]。例如，缺乏许多中转域汇聚在一起的交换等功能。有没有“更好”的技术来生成旨在为Internet建模的拓扑？更根本的是，应该如何评估一种拓扑生成技术(即，如何“更好”地衡量)？2)拓扑缩放。尽管在支持大规模模拟方面取得了很大进展[33]，但大多数研究人员将继续在比目标运行的大规模网络更小的拓扑上模拟他们的协议。应该如何配置较小的拓扑，以使它们合理地反映较大的对应拓扑？是否有一种拓扑缩放理论可以为配置各种大小的拓扑提供基本依据？3)拓扑用途。PI对拓扑建模的主要兴趣是为大规模仿真提供基础。然而，促进在模拟中使用拓扑必须不仅仅是提供理论上合理的模型，还必须包括一组用于图形可视化、路由表构建等的补充工具。哪些可视化工具对研究人员有用并有助于准确直观地了解底层拓扑？如何在构建路由表时有效地反映不同的路由策略？研究人员建议(1)在拓扑建模领域解决这些和其他基本问题，(2)在向广大研究社区提供的一套拓扑工具和基准中反映他们的理解。这项工作将建立在PI先前在网际网络建模方面的经验基础上。拟议的工作将有助于对拓扑建模领域的基本理解。这项工作将包括一套评估标准，以评估拓扑生成方法的质量和拓扑模型的改进。这项工作还将产生拓扑缩放的进化理论，对于使用比目标更小的拓扑进行高效模拟具有重要意义。除了对基本理解作出贡献外，拟议工作的一个核心组成部分是一套工具和基准，将沿袭GT-ITM套件的传统向整个研究界提供。这些工具将允许其他研究人员生成拓扑，评估候选拓扑建模方法的质量，利用基于当前和未来技术的基准，并与拓扑可视化交互。

Network simulation is an indispensable tool for researchers seeking to understand the principles of network architecture and protocol design. A key parameter in any moderate to large-scale simulation is the topology, i.e., the way the nodes of the network are organized and connected to each other. "Good" models for topology are essential for good simulations. The PIs have developed graph modeling software that currently is widely used as a tool for generating topologies, particularly models of large internetworks. The Georgia Tech Internet Topology Models (GT-ITM) package allows researchers to construct model topologies whose structure arguably resembles the node-level structure of the Internet: routers or switches, connected by (bidirectional) links, and grouped into domains. The GT-ITM software is included with "ns2" [2], the defacto open-source standard for network simulation. Despite the wide-spread use of GT-ITM, in general, and its transit-stub model, in particular, a number of critical and fundamental questions remain unanswered about network topology modeling. For example, 1)Topology models. Recent data indicates that the current Internet topology has some properties that are not well reflected in the transit-stub model of GT-ITM [17]. For example, features such as the exchanges where many transit domains come together are lacking. Are there "better" techniques to generate topologies intended to model the Internet? More fundamentally, how should a topology generation technique be evaluated (i.e., how is "better" measured)? 2)Topology scaling. Although strides are being made in supporting large-scale simulations [33], most researchers will continue to simulate their protocols on topologies that are smaller than the target operational large-scale networks. How should smaller topologies be configured so that they reasonable reflect their larger counterparts? Is there a theory of topology scaling that can provide the fundamental grounding for configuring topologies of various sizes? 3)Topology use. The PIs primary interest in topology modeling is to provide a foundation for large-scale simulations. Facilitating the use of topologies in simulations must go beyond providing theoretically sound models, however, and include a set of complementary tools for graph visualization, routing table construction, etc. What visualization tools are useful to researchers and assist in accurate intuitive understanding of underlying topology? How can different routing policies be effectively reflected in routing table construction? The researchers propose (1) to address these and other fundamental questions in the area of topology modeling and (2) to reflect their understanding in a set of topology tools and benchmarks made available to the research community at large. This work will build on the PIs prior experience in modeling internetworks. The proposed work will contribute to fundamental understanding in the area of topology modeling. The work will include a set of evaluation criteria to assess the quality of a topology generation method and improvements in topology models. The work will also produce an evolutionary theory of topology scaling, with implications for efficient simulation using topologies that are smaller than the target. In addition to contributions to fundamental understanding, a central component of the proposed work is a set of tools and benchmarks to be made available to the research community at large, following in the tradition of the GT-ITMsuite. These tools will allow other researchers to generate topologies, assess the quality of candidate topology modeling methods, utilize benchmarks based on current and future technologies, and interact with a visualization of topology.