多尺度现象和大规模并行处理

Multi-Scale Phenomena and Large-Scale Computation

目前在超级计算应用上主要有两方面的瓶颈问题,一是计算的结果的可靠性和精确性,这主要与物理模型和计算精度相关;二是实际应用的计算规模较小,当需要对物理模型进行精细计算的时候,由于计算效率低,无法把机器的计算能力有效地转化为计算规模的扩大。为了解决这些问题,一般而言,需要从建立物理模型开始考虑并行性的问题。多尺度现象是自然界的普遍现象。不同尺度之间的相互关联,即从小尺度下的性质如何获得大尺度下的性质,它的过程细节是什么,这直接影响到人们对复杂系统发展演化规律的深入认识,这就是介尺度的问题。如何通过认识介尺度过程而建立起大、小尺度的相互联系是科学家和工程科技人员面临的共同挑战。而这也正是超级计算机强大的处理能力的用武之地。最后,文章提出介尺度理论和算法应当是计算科学和技术中值得重视的重要方向之一。

In supercomputing applications there exit two major bottleneck problems: the first is reliability and accuracy, which is related mainly with its physical model and computational fineness; the other is the difficulty in achieving large scale computation, which prevents efficient transformation of the computing power to the capability of enlarging the computational scale because of the low efficiency of large scale parallel computation. This problem results in that the real applications with very large scale cannot be run well or even don't work on the supercomputers. To solve these problems, in general, one needs to consider the parallelism begin with establishment of the physical models. The multi-scale phenomena are universal in the nature. It is critical for the people to understand deeply the rules of movement and evolution of the complex systems, to know how the properties in a larger scale are acquired from those in a smaller scale, and what sophisticated process happens in between. This is called 'meso-scale' problem. It is a challenge for scientists and engineers to discover the relationship and the law of transition between the small scale and the large scale by understanding the meso-scale. There is a tremendous demand for supercomputing power in order to address the meso-scale issue. This paper finally points out that the theory and algorithm of computation for meso-scale are one of the most important research directions in computational science and technology.