ITR: Designing a Programming Environment for Network Systems

为有效处理数据包而设计的网络系统必须满足两个要求：(1)支持大量的高带宽链路，从而支持大的系统吞吐量；(2)提供广泛的服务（如VPN、入侵检测、病毒扫描、差异化服务和覆盖网络处理）。为了同时满足这些需求，出现了一种新的处理器，称为网络处理器（npu）。网络处理器与通用处理器、特定功能的协处理器和其他可重构逻辑元件相结合，为设计网络系统创造了一个强大的平台。不幸的是，编程这些嵌入式、高度并行的网络系统所需的方法和工具还处于起步阶段。今天，要在这样的系统上开发高性能的包处理应用程序，程序员必须非常熟悉硬件。程序员需要在设计时在不同的处理器类型之间划分他们的应用程序，并开发手动调优的代码来管理网络系统资源。这种方法冗长、缓慢，并且产生的代码难以从一个基于npu的平台移植到另一个平台。随着下一代网络系统和npu支持更多数量的处理器，更多样化的内存层次结构和更复杂的处理器互连，编程网络系统的难度只会随着时间的推移而增加。这项研究的目标是创建一个编程环境，使未来的网络系统像今天的工作站和服务器一样容易编程。实现这一愿景需要一个跨网络、编程语言、编译器、运行时系统和软件工程领域的跨学科研究计划。提议的多学科活动将产生一个编程环境，其中包括：(1)用于指定包处理应用程序的特定领域编程语言；(2)一套分析应用程序运行时特征的工具；(3)一套编译时工具和技术，用于将包处理应用程序映射到复杂的网络系统架构；(4)一个动态调整资源分配的运行时系统，以创建对攻击具有鲁棒性的网络系统。总之，这组工具和方法将创造一个环境，促进可编程网络系统上便携、高性能数据包处理应用程序的快速开发，这是未来几代多业务信息网络的基石。

Network systems-designed to process packets efficiently-must meet two requirements: (1) support a large number of high-bandwidth links, and hence large system throughputs; and (2) offer a wide range of services (such as VPN, intrusion detection, virus scanning, differentiated services, and overlay network processing). To meet these requirements simultaneously, a new breed of processors, referred to as network processors (NPUs), has emerged. Network processors, when combined with general-purpose processors, function-specific coprocessors and other reconfigurable logic elements, create a powerful platform for designing network systems. Unfortunately, the methodologies and tools needed for programming these embedded, highly-parallel network systems are in their infancy. Today, to develop high-performance packet processing applications on such systems, programmers are required to be intimately familiar with the hardware. Programmers are required to partition their application at design time among the different processor types and develop hand-tuned code for managing network system resources. This approach is tedious, slow, and yields code that is difficult to port from one NPU-based platform to another. With the next-generation of network systems and NPUs supporting a larger number of processors, more diverse memory hierarchies, and more complex processor interconnects, the difficulty in programming network systems will only increase over time. The objective of the proposed research is to create a programming environment that will make future generations of network systems as easily programmable as today's workstations and servers. Realizing this vision requires an inter-disciplinary research program that spans the fields of networking, programming languages, compilers, run-time systems, and software engineering. The proposed multi-disciplinary activity will yield a programming environment that includes: (1) a domain-specific programming language for specifying packet processing applications; (2) a set of tools for profiling run-time characteristics of applications; (3) a set of compile-time tools and techniques for mapping packet processing applications onto complex network system architectures; and (4) a run-time system that adapts resource allocations dynamically to create network systems that are robust against attacks. Together, this set of tools and methodologies will create an environment that will facilitate rapid development of portable, high-performance packet processing applications on programmable network systems-the building blocks of future generations of multi-service information networks.