Protocols and Algorithms for Large-scale Multi-Substrate Networks

网络设备数量的不断增加以及设备同时连接到多个通信网络的能力的出现为新的网络体系结构奠定了基础。机器对机器通信、内容感知网络和网络物理系统的各种专用网络的互连创建了异构网络的集合。在传统的以互联网为中心的体系结构中，所有网络都连接到互联网，但彼此之间不连接。然而，在许多网络都是移动的并且只有间歇性连接的环境中，对互联网永久连接的假设是有限的。这就产生了对另一种网络设计的需求，在这种设计中，互联网连接到一些网络，但不是所有网络。我们提出了一个关于异构网络（称为基板网络）的非常大的集合的协议和算法的研究议程，其中设备具有多个基板网络的附件。我们寻求支持几乎无限数量的任意大小的基板网络，由于移动性或故障而动态改变基板连接。我们建议解决能够实现这一愿景的关键问题。我们将研究大型互联网中的路由方法，由于网络动态和移动性，一致的全局路由表是不可行的。我们将开发算法，使具有多个网络接口的设备能够选择一个或多个基板网络进行数据传输，同时满足稳定性标准。研究方法包括算法设计、协议设计与验证、大规模仿真、分析性能评估和协议实现。一个单独的推力，但我们的研究议程的一个组成部分，是分析方法的发展，以评估多基板网络的延迟和吞吐量性能。基于我们之前对随机网络演算的研究，我们将致力于创建一个基础理论来解释统计复用网络中的带宽可用性和延迟。我们将解决网络随机最小加系统理论中长期存在的开放性问题。特别地，我们建议在网络微积分的系统理论中发展具有反馈的随机系统评估的分析方法。我们也试图找到改进的下界性能指标推导与随机网络演算。

The ever increasing number of networked devices and the emerging ability of devices to simultaneously connect to multiple communication networks lays the foundation for a new networking architecture. The interconnection of diverse specialized networks for machine-to-machine communications, content-aware networking, and cyber-physical systems creates a collection of heterogeneous networks. In a traditional Internet-centric architecture, all networks connect to the Internet, but not to each other. However, in an environment where many networks are mobile and have only intermittent connectivity, the assumption of permanent connectivity to the Internet is limiting. This creates the need for an alternative network design, where the Internet is connected to some but not all networks. We propose a research agenda on protocols and algorithms for very large collections of heterogeneous networks, referred to as substrate networks, where devices have attachments to multiple substrate networks. We seek to support a virtually unlimited number of substrate networks of arbitrary sizes, with dynamic changes to substrate connectivity due to mobility or failures. We propose to address key problems that can realize this vision. We will investigate methods for routing in large internetworks, where consistent global routing tables are not viable due to network dynamics and mobility. We will develop algorithms that enable devices with multiple network interfaces to select one or more substrate networks for data transmission, while satisfying stability criteria. The research methods include algorithm design, protocol design and validation, large scale simulations, analytical performance evaluation, and protocol implementation. A separate thrust, but an integral part of our research agenda, is the development of analytical methods to evaluate the delay and throughput performance of multi-substrate networks. Building on our prior research on the stochastic network calculus, we will work towards the creation of a foundational theory to reason about bandwidth availability and delays in networks with statistical multiplexing. We will address long-standing open problems in the stochastic min-plus system theory for networks. In particular, we propose to develop analytical methods for the evaluation of random systems with feedback within the systems theory of the network calculus. We also seek to find improved lower bounds for performance metrics derived with the stochastic network calculus.