An approach to increase reliability of HPC simulation, application to the Gysela5D code

Abstract

Abstract. Reproducibility of results is a strong requirement in most fields of research for experimental results to be called science. For results obtained through simulation software using high performance computing (HPC) this translates as code quality requirements. While there are many works focusing on software quality, these typically do not take the specificities of HPC scientific simulation software into account. This paper presents an approach to introduce quality procedures in HPC scientific simulation software while remaining the less invasive as possible so as to ease its adoption. The approach relies on quality procedures including human code review and automated testing and offers a dedicated procedure to help correct defects found this way. These procedures are integrated in a development work-flow designed to improve the traceability of defects. By implementing this approach for the development of the Gysela code, we show that it is indeed viable and that the return on investment is positive. We also identify multiple reusable elements developed for this experiment that should reduce the cost of adopting the approach for other codes as well as some aspects that can still be improved to ensure a widespread propagation of the approach in the community.

Building and Auto-Tuning Computing Kernels: Experimenting with Boast and Starpu in the Gysela Code

“…The optimization process can thus be executed independently of the compilation, at a much lower frequency, typically only when the kernel has been modified or the machine hardware and software stack has changed. An interesting improvement of this process would be to automate it by using the already existing gysela continuous integration platform [6].…”

“…The code for the Lagrange 1D kernel is in C. It's made of three functions: fn_init, which initializes the input data; fn_advec, which is the kernel; and fn_verif, which verifies the correctness of the output. The source code for this kernel, with the auto-tuning code, is available in the BOAST documentation 6 .…”

Design and Study of Elastic Recovery in HPC Applications

“…The methods described in this article will be applicable in numerous categories, i.a., cosmological applications, climate simulations, and reanalyses. For instance, GISELA5D, a plasma simulation for fusion reactions, already uses the HDF5 format to combine resiliency and analysis [5].…”