SHF: EAGER: Developing General Techniques for Tightening Bounds of the Data-Movement Complexity of Large Scale Parallel Applications

University Of Colorado At Denver-Downtown Campus, Denver CO

Investigators

Abstract

Enabling faster and more energy-efficient numerical simulations is critical, for example, in basic science for enabling novel scientific discoveries, or in engineering for developing revolutionary new products. In the last decades, technology trends in computer systems have resulted in widely differing rates of improvement in computational throughput as compared to data movement performance. With future systems, the cost of data movement through the memory hierarchy is expected to become even more dominant relative to the cost of performing arithmetic operations, both in terms of time and energy. Consequently, the data movement and communication costs of a numerical simulation become determinant factors for the time to solution and the energy consumption. This research is developing novel generic techniques for tightening bounds on the data movement complexity of numerical simulations. The outcomes of this research are methods to derive the communication needs of numerical simulations. The ability to assess the optimality of an algorithm enables understanding of the implications of various computing platform's parameters on the performance of a numerical simulation. The PIs are developing novel generic techniques for tightening bounds on the data movement complexity of a given algorithm on a given architecture. This work engages in novel interdisciplinary perspectives and brings together applied mathematics, theoretical computer science, data analytics and high performance computing.