SBIR Phase II: Ultra-large and low-cost Electrodynamic Modeling in Commercial Clouds

这个小企业创新研究（SBIR）第二阶段项目将为商业云中的电动力学模拟开发新的算法。在光学和射频器件的计算原型中，电动力学仿真是必不可少的。当前的商业解决方案需要在软件和硬件方面进行大量的前期投资。由于有限的内部计算资源，它们的生产率也很低。该项目将开发一种基于云的模拟服务，该服务将提供按需、按使用付费、几乎无限的模拟功能，而无需用户购买任何硬件或软件。这种仿真技术将大大提高生产效率，降低进入门槛，并在光学和射频应用的设计中最大限度地降低成本。此外，它将服务于多个科学和工程学科，并将促进在科学发现和工程设计的所有阶段使用高性能仿真。它将彻底改变大规模的科学模拟，并允许任何有互联网接入的人获得前所未有的模拟能力。最终，这项技术将为工程师和科学家提供直观的、按需的、几乎无限可扩展的科学软件，从而帮助加速健康、能源和国防行业的研发，其价格远低于任何其他云计算服务。目前还没有能够在商业云中高效运行的电磁仿真软件，因为它们是在底层计算平台同质化、cpu间通信时间低（即低延迟）的假设下编写的。从最基本的时空离散化开始，我们将开发第一个延迟容忍和云优化的电动力学仿真软件。第一阶段完成的工作开发了几个基于云的概念验证型电动求解器，并证明了商业云计算平台是提供极低成本计算的可行选择。第二阶段开发的通用求解器将服务于广泛的应用，并将作为云中的电磁仿真服务商业化，具有完全沉浸式的基于web的可视化界面。

This Small Business Innovation Research (SBIR) Phase II project will develop novel algorithms for electrodynamic simulations in commercial clouds. Electrodynamic simulation is essential for computational prototyping in optical and radio-frequency devices. Current commercial solutions require significant upfront investment of both software and hardware. They also suffer from low productivity due to limited in-house computing resources. This project will develop a cloud-based simulation service which will provide on-demand, pay-per-use, virtually unlimited simulation capability without requiring users to purchase any hardware or software. This simulation technology will greatly improve productivity, reduce the barrier to entry, and minimize cost in the design of optical and radio-frequency applications. Moreover, it will serve multiple scientific and engineering disciplines and will facilitate the use of high-performance simulation in all phases of scientific discovery and engineering design. It will revolutionize large-scale scientific simulation and allow anyone with internet access to gain unprecedented simulation power. Ultimately, this technology will help accelerate research and development in the health, energy, and defense industries by offering intuitive, on-demand, and practically infinitely scalable scientific software to engineers and scientists at a price-point significantly lower than any other cloud computing service.There is currently no electromagnetic simulation software which performs efficiently in commercial clouds, because they were written under the assumption that the underlying computing platform is homogenous and has low inter-CPU communication time, i.e. low latency. Starting from the most fundamental level of space-time discretization, we will develop the first electrodynamic simulation software that is latency-tolerant and cloud-optimized. The work completed in Phase I developed several proof-of-concept cloud-based electrodynamic solvers and demonstrated that commercial cloud computing platform is a viable option for offering extremely low cost computing. The general solver to be developed in Phase II will serve a broad set of applications, and will be commercialized as the electromagnetic simulation service in the cloud, featuring a fully immersive web-based visualization interface.