AitF: Theory and Practice of Probabilistic Network Programming

许多计算机网络中的一个共同目标是平衡多条路径上的流量，以便提高吞吐量、减少拥塞并实现更高的利用率。尽管有大量的理论文献说明如何通过容量有限的网络最好地路由流量，但现在的大多数网络都依赖于易于实施但在实践中并不总是表现良好的方案。这在一定程度上是由于研究高级路由方案的理论研究人员社区与设计、构建和运营网络的系统研究人员社区之间的差距。这也是由于缺乏高级编程语言，使得在网络中编程高级路由功能变得容易，并对此类程序的性能进行推理。本项目将通过开发可以在现有网络硬件上实现的不经意和半不经意的路由算法，并伴随着概率网络编程语言和编译器来推进路由的理论和实践。这项研究的智能优势在于开发了新的路由算法，该算法针对最小化拥塞和平衡负载等目标进行了优化，设计了基于概率语义的新编程语言，可以用来表达和推理这些算法，并构建了可以部署在软件定义的网络平台上的原型实现。该项目将通过开发算法和工具对社会产生广泛影响，这些算法和工具有助于使网络更加可靠和高效，同时简化网络管理。为了加强这一广泛的影响，该项目包括在算法和网络的交叉点组织一次跨学科研讨会。

A common objective in many computer networks is to balance traffic across multiple paths in order to improve throughput, reduce congestion, and achieve higher utilization. Although there is an extensive theoretical literature on how best to route traffic through a capacitated network, most networks today rely on schemes that are easy to implement but do not always perform well in practice. This is due, in part, to the gap between the community of theory researchers who study advanced routing schemes and the community of systems researchers who design, build, and operate networks. It is also partly due to the lack of high-level programming languages that make it easy to program advanced routing functionalities in networks and to reason about the performance of such programs.This project will advance the theory and practice of routing by developing oblivious and semi-oblivious routing algorithms that can be implemented on top of current network hardware, accompanied by a probabilistic network programming language and compiler. The intellectual merit of this research lies in developing new routing algorithms that optimize for objectives such as minimizing congestion and balancing load, designing new programming languages based on probabilistic semantics that can be used to express and reason about these algorithms, and building prototype implementations that can be deployed on software-defined networking platforms. This project will have broad impact on society by developing algorithms and tools that could help make networks more reliable and efficient while simplifying network management. To reinforce this broad impact, the project incorporates organizing an interdisciplinary workshop at the intersection between algorithms and networking.