Enhancing Energy sector efficiency: A study on supercomputer performance in optimizing energy systems

Abstract

The paper discusses the question of how to evaluate actual performance of a supercomputer system, which is expected on solution to real scientific or engineering problem. In practice this actual performance is rather far from the rather well-known peak performance. In particular, the paper contains a review of the computational problems most commonly solved on petaflops supercomputers as well as the corresponding methods designed for exaflops supercomputers. A proposed technique for measuring the characteristics of the supercomputer using a Particle-In-Cell (PIC) code is described. The choice of this particular numerical method is based on its features, namely that PIC method involves a wide variety of numerical techniques, and thus it is very hard to optimize. The analysis of scalability, parallel efficiency and acceleration rate, which is possible at a particular supercomputer is presented, as well as the analysis of the performance of multi-architecture supercomputer nodes. The integrated criteria to evaluate the real performance of supercomputer system is proposed.