CAREER: Large-scale Dynamic Reconfigurable Networks

该奖项全部或部分由《2021年美国救援计划法案》（公法117-2）资助。由于社会对在线服务的依赖日益增加，底层通信网络的可靠工作至关重要。对于现代在线服务，新出现的工作负载（如远程视频呼叫、增强现实、机器学习和医疗保健）高度依赖底层网络的响应。然而，当今网络基础设施的设计仍然将网络的物理层视为具有最小可重构性的静态黑盒。该项目旨在开发适用于数据中心网络和软件定义专用广域网的大规模动态可重构网络的新范例，以改善服务交付。该项目的核心任务是使物理层可重构性成为未来网络的内在组成部分。该项目侧重于高影响力用例和应用，通过利用光学技术为可重构网络开发新颖的解决方案。为了使大规模可重构网络成为现实，本提案解决了高性能可重构系统的基本挑战，包括：(1)一套算法和系统设计技术，用于联合数据中心网络中新兴分布式机器学习作业的并行化策略来共同优化网络拓扑。(2)一套基于优化和学习的技术，为可重构软件定义专用广域网构建实用的跨层解决方案，同时保证性能。(3)平衡可重构系统的算法和工程基础的技术。部署可重构网络将使世界各地的用户能够获得可靠和快速的在线服务。因此，该项目具有很高的行业影响潜力。从教育的角度来看，该项目将开发一门新的研究生课程，关于机器学习系统和机器学习系统。这个机器学习和光学系统交叉的新兴领域是由人工智能的各种应用的爆炸式增长和大规模系统的复杂性驱动的。该项目将开发各种模拟和仿真环境，重点是机器学习工作负载和技术，这些环境将供大量可能在这些领域没有专业知识的学生和研究人员使用。通过本项目工作产生的数据将包括论文、源代码和基准，并将在以下网站上发布：http://reconfignets.csail.mit.edu/数据将在奖励期后至少保留三年。该奖项反映了美国国家科学基金会的法定使命，并通过使用基金会的知识价值和更广泛的影响审查标准进行评估，被认为值得支持。

This award is funded in whole or in part under the American Rescue Plan Act of 2021 (Public Law 117-2).Due to society’s ever-increasing dependency on online services, reliable working of the underlying communication networks is of paramount importance. For modern online services, emerging workloads (such as remote video calls, augmented reality, machine learning, and health care) depend highly on the underlying network’s response. However, the design of today’s network infrastructures still treats the physical layer of networks as a static black box with minimal reconfigurability. This project seeks to develop new paradigms for large-scale dynamic reconfigurable networks that are applicable to datacenter networks and software-defined private wide area networks to improve service delivery. The core mission of the project is to make physical layer reconfigurability an intrinsic part of future networks. The project focuses on high-impact use-cases and applications to develop novel solutions for reconfigurable networks by leveraging optical technologies. To make large-scale reconfigurable networks a reality, this proposal tackles the foundational challenges of high-performance reconfigurable systems, including: (1) A set of algorithmic and system design techniques to co-optimize the network topology jointly with the parallelization strategy of emerging distributed machine learning jobs in datacenter networks. (2) A set of optimization and learning-based techniques to build practical cross-layer solutions for reconfigurable software-defined private wide-area networks while providing guaranteed performance. (3) Techniques to balance algorithmic and engineering foundations for reconfigurable systems. Deploying reconfigurable networks will enable users around the world to have access to reliable and fast online services. As a result, this project has the potential of high industry impact. From an educational perspective, the project will develop a new graduate-level course on Systems for Machine Learning and Machine Learning for Systems. This emerging area at the intersection of machine learning and optical systems is driven by the explosive growth of diverse applications of artificial intelligence and the complexity of large-scale systems. This project will develop a variety of simulated and emulated environments with a focus on machine learning workloads and techniques which will be accessible to a large community of students and researchers who may not have expertise in these areas.The data generated through the work in this project will consist of papers, source code, and benchmarks and will be released at the following website: http://reconfignets.csail.mit.edu/ Data will be retained for at least three years beyond the award period.This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.