An Event-driven Programmable Network Architecture for the Next Generation Internet

近年来，使用Internet的应用程序的数量和多样性呈爆炸式增长。过去，互联网上的大部分数据传输都是可靠的、有序的、非实时的，而且是在一对固定的主机之间进行的。然而，下一代互联网将需要支持非常多样化的环境（例如，异构的有线/无线和移动网络）、应用程序（例如，多媒体、WWW、telnet）和服务（多播、不同级别的服务质量等）。由于具有不同需求的应用程序需要对网络的动态做出不同的反应，因此下一代Internet协议必须在处理特定于应用程序的需求方面更加主动。此外，还需要为新兴应用程序开发和快速部署新的服务。在这个项目中，我们提出了一种新颖的事件驱动可编程网络体系结构，它具有可扩展性、鲁棒性、灵活性、可扩展性和易于编程性。我们的架构支持用户和服务提供商对新网络服务的动态实例化、虚拟网络的动态创建、服务质量、服务广告、灵活的网络资源管理以及竞争应用程序之间的资源仲裁。我们的目标是使网络编程几乎像单机编程一样简单。我们的方法是提供网络的事件驱动视图。“虚拟网络”由一组服务定义，每个服务都是一个事件、事件处理程序对。网络提供了一组默认的服务。每个服务提供商都可以创建自己的“虚拟网络上下文”，这是虚拟网络的实例。虚拟网络上下文具有层次结构，即，服务提供者可以允许用户或应用程序在服务提供者的上下文中创建自己的虚拟网络上下文。每个虚拟网络上下文都继承其父上下文的服务，但可以自定义这些服务或创建新服务。可以通过将事件绑定到不同的事件处理程序或创建新事件（可以由其他事件处理程序触发）来定制服务。网络范围内的“分布式服务目录”维护可用的和公开分发的服务的（部分）库。在此模型中，只需更改与用户相关的少数服务，就可以轻松地定制虚拟网络上下文。同时，通过混合和匹配公共可用服务的菜单，可以很容易地组合新服务和构建整个虚拟网络。为了使编程服务变得简单，我们的体系结构为应用程序提供了分布式网络状态的简单编程抽象，从而降低了开发服务的复杂性。分布式网络管理器提供简单的机制，用于访问、更新、缓存和锁定属于潜在异构网络中的多个节点的本地和非本地网络状态。网络管理器是基于我们过去开发的“骨干”基础设施的自配置分布式软件，它还提供顶级虚拟网络上下文之间的资源管理、资源分配和资源仲裁。我们相信我们的组合服务模型和网络管理架构在当代研究中是独一无二的，并且可能被证明是下一代互联网的合适架构。为了更好地理解真实环境中架构和服务需求的微妙之处，我们将构建一个由横跨全国的异构有线/无线网络组成的四个集群测试平台，集群位于香槟、芝加哥、波士顿和圣何塞。我们将与摩托罗拉、TranSwitch和Maverick半导体公司合作建立基础设施。该环境的目标是为移动计算环境中的单播和多播流提供高质量的服务。我们之所以选择这个应用程序环境，是因为它的复杂性以及我们以前在传统Internet上下文中构建此类环境的专业知识。我们相信，在这个项目的需要，我们将能够开发一个架构，可以有效地解决下一代互联网的多样化和复杂的需求。

Recent years have witnessed an explosive growth in the volume and the diversity of applications that use the Internet. IN the past, most of the data traffic traversing the Internet was reliable, sequenced, non real-time, and between a stationary pair of hosts. However, the next generation Internet will need to support very diverse environments (e.g., heterogeneous wireline/wireless and mobile networks), applications (e.g., multimedia, WWW, telnet), and services (multicast, different classes of quality of service, etc.). Because applications with diverse requirements need to react differently to the dynamics of the network, the next generation Internet protocols will, of necessity, have to be more proactive in the way they deal with application-specific requirements. Additionally, new services will need to be developed and rapidly deployed for emerging applications. In this project, we propose a novel event-drive programmable network architecture that is scalable, robust, flexible, extensible, and simple to program. Our architecture enables on-the-fly instantiation of new network services both by users and service providers, dynamic creation of virtual networks, quality of service, service advertisement, flexible network resource management, and resource arbitration among contending applications. Our goal is to make programming the network almost as simple as programming a stand-alone computer. Our approach is to provide an event-drive view of the network. A 'virtual network' is defined by a set of services, each of which is an event, event-handler pair. The network provides a default set of services. Each service provider can create its own 'virtual network context', which is an instantiation of a virtual network. Virtual network contexts have a hierarchical structure, i.e., a service provide may allow users or applications to create their own virtual network contexts within the context of the service provide. Each virtual network context inherits the services of its parent context, but can customize these services or create new ones. Services can be customized either by binding an event to a different event-handler, or by creating new events (which can be triggered by other event-handlers). A network-wide 'distributed services directory' maintains a (partial) library of available and publicly disseminated services. In this model, a virtual network context can be easily customized piecemeal by changing only a few of the services relevant to a user. At the same time, it is easy to compose new services and build entire virtual networks by mixing and matching from a menu of publicly available services. In order to make programming services simple, our architecture provides a simple programming abstraction of the distributed network state to applications, thereby reducing the complexity of developing services. A distributed network manager provides simple mechanisms for accessing, updating, caching, and locking both local and non-local network state belonging to multiple nodes in potentially heterogeneous networks. The network manager is a self-configuring distributed software based on the 'spine' infrastructure that we have developed in the past, and also provides resource management, resource allocation, and resource arbitration among top-level virtual network contexts. We believe that our combined service model and network management architecture is unique in contemporary research, and may prove to be an appropriate architecture for the next generation Internet. In order to better understand the subtleties of the architecture and service requirements in a real environment, we will build a four cluster testbed consisting of heterogeneous wireline/wireless networks spanning the country, with clusters located in Champaign, Chicago, Boston, and San Jose. We will build hte infrastructure in collaboration iwth Motorola, TranSwitch, and Maverick Semiconductor. The goal of htis evnvironment is to provide quality of service for unicast and multicast flows in a mobile computing environment. We choose this application environment because of its comcplexity and our previous expertise in building such environments in the context of hte traditional Internet. We believe that at the ned of this project, we iwll be able to develop an architecture that can effectively address the diverse and complex requirements of hte next generation Internet.