NeTS: Small: Flange: A Domain Specific Language for Network

越来越多的软件依赖于网络，与此同时，网络本身也通过OpenFlow等技术变得更加可编程，为网络效率和网络内计算提供了广泛的机会。虽然新的编程模型试图使网络重新配置更容易，但没有一个提供工具来解决软件控制路由和网络内计算服务放置的联合问题。该项目将设计和构建一种称为法兰的领域特定语言（DSL），用于管理网络中的服务。该项目的方法是在网络中编组可配置的转发和计算设备，并进行普遍监控以支持服务适应。该项目包括构建和发布一个原型编译器、运行时、系统和前端，使法兰能够使用专用语法（独立DSL）和嵌入宿主语言进行编程。该项目的成功标准是，使用原型的网络管理员能够使用不到一页的法兰代码设置新服务（例如视频转码）。使用相同的机制，网络工程师应该能够启动一个持久的功能来监视网络状况，并通过将流量重定向到备用路径来响应饱和。这个项目的新奇之处是双重的。首先，该项目的语义基础将网络的硬件和软件方面统一对待，构成一个单一的过程网络。例如，该项目将转发建模为网络内服务，同时将其与网络中其他可能实例化的服务（如转码或压缩）放在相同的角度考虑。其次，法兰语言将为大型网络提供新颖的全局视图编程抽象。法兰程序将能够检查网络实体，聚合信息，信号和响应事件，所有这些都不需要参考它们自己的执行地点（通常是物理分布的）。这里提出的工作将影响以国家科学为重点的网络基础设施以及商业应用的网络运行效率和灵活性，在商业应用中，网络计算的经济影响是显而易见的。此外，基础法兰的设计决策与教学相关，并将在pi教授的课程中呈现。

An increasing amount of software depends on the network, and at the same time networks themselves are becoming more programmable via technologies like OpenFlow, opening a wide range of opportunities both for network efficiencies and in-network computation. While new programming models seek to make network reconfiguration easier, none provide tools to solve the joint problem of software-controlled routing and placement of in-network computational services. This project will design and build a domain specific language (DSL), called Flange, for managing services in networks. The project's approach marshalls configurable forwarding and computing devices in the network, with pervasive monitoring to support service adaptation. The project includes building and releasing a prototype compiler, runtime, system and front-ends that enable Flange programming both with dedicated syntax (standalone DSL) and embedded in host languages. The success criteria for this project is that a network administrator using the prototype is able to set a new service (e.g. video-transcoding) using less than a page of Flange code. Using the same mechanism, a network engineer should be able to launch a persistent function that monitors network conditions and responds to saturation by redirecting traffic to an alternate path. The novelty of the project is twofold. First, the semantic foundation at the root of the project treats both hardware and software aspects of networks uniformly, as constituting a single process network. For example, the project models forwarding as an in-the-network service, while considering it in the same light as other potentially instantiated services in the network such as transcoding or compression. Second, the Flange language will provide novel global view programming abstractions for large networks. Flange programs will be able to inspect network entities, aggregate information, signal and respond to events, all without reference to their own place of execution (which is often physically distributed). The work proposed here stands to impact network operational efficiency and flexibility for national science-focused cyberinfrastructure as well as in commercial applications, where the economic impact of networked computing is clear. In addition, the design decisions the underly Flange are pedagogically relevant and will be presented in courses taught by the PIs.