Collaborative Research: Advanced Parallel Computing Techniques with Applications to Computational Cosmology

拥有超过10万个处理器的并行机器正在建造中。美国国家科学基金会广泛使用的TeraScale设施已经部署了3000个处理器的机器。算法的进步使得以更快的速度解决问题成为可能。然而，这些算法的复杂性和并行化仍然是阻碍科学计算进步的一个障碍。该项目将开发一种基于对象的方法来简化高效并行应用程序的开发。将开发一个运行时系统来自动地将用户的基于对象的应用程序分布在不同的处理器之间。该系统对运行中的程序进行动态观察，并动态应用一套智能策略进行优化。并行组件可以插入和退出正在运行的计算，并以灵活的方式相互交换数据。这种基础设施的发展将提供一个框架，使构建面向粒子的并行程序变得容易，用于模拟计算天体物理学中的星系、暗物质、恒星、行星和气体粒子。新的框架将允许模拟在数十亿个粒子和数万个处理器的情况下高效运行。这些模拟将能够与卫星和地面天文数据直接接触，以获得高精度的宇宙学参数，并研究行星如何形成。

Parallel machines with over a hundred thousand processors are now being built. NSF's widely accessible TeraScale facilities have already deployed a 3,000 processor machine. Algorithmic advances have made it possible to solve problems at a much faster rate. However, complexity of these algorithms and parallelizing them remain a hindrance to advances in scientific computing. This project will develop an object based methodology to simplify development of highly efficient parallel applications. A run-time system will be developed to automatically distribute the user's object based application among various processors. This novel system observes the running program to apply dynamically an intelligent suite of strategies to effect optimization. Parallel components can be plugged in and out of running computations and exchange data with each other in a flexible manner. The development of this infrastructure will provide a framework to allow easy building of particle-oriented parallel programs to be used to model galaxies, dark matter, stars, planets, and gas particles in computational astrophysics. The new framework will allow simulations to operate efficiently with billions of particles and tens of thousands of processors. These simulations will be able to make direct contact with satellite and ground-based astronomical data to obtain cosmological parameters with high precision and to study how planets form.