It will be shown a course targeted to improve the overall performance of a code in Solid Earth, currently in use at the CHEESE Center of Excellence H2020 project.
Starting from a description of main architectures characterizing the HPC world,
parallel performance profiling tools will be provided on examples of the code to find the so-called performance bottlenecks. Starting from the profiling analysis, it will show how and where to intervene with respect to the hardware characterization of the HPC machine used for the investigation to improve the overall performance of serial and parallel codes.
Particular attention will be given to tools and techniques to improve the performance of the codes with respect to I / O and vectorization.
Skills:
At the end of the course the student will be able to:
use a concrete methodology to improve the performance of a code in Solid Earth already in use in the context of ChEESE project
find and solve all the main bottlenecks of an application with respect to appropriate computational metrics and the machine used
Target audience:
Researchers in Solid Earth interested to learn and use all those techniques and related tools that may allow them to improve the performance of their code on current HPC architectures in the shortest possible time.
Pre-requisites:
Basic knowledge of LINUX/UNIX.
Knowledge of C, FORTRAN, MPI or OpenMP is recommended. Notions of Parallel computing techniques and algorithms for Solid Earth applications
Coordinating Teachers: Dr. V Ruggiero, G.Amati.