We have developed software that enables running (but is not restricted to) MPI-based parallel applications given as task graphs onto parallel and distributed HPC systems. This software enables executing large scale parallel and distributed HPC applications in a drastically reduced form, obtaining execution traces in a reasonable time. Runtime observations can be used to improve the future runs of this application on a real system. GRAPPLER generates application execution traces per processing unit and per type of load (computation and communication between tasks) in a human-readable format which helps identify bottlenecks of the application, the platform, and the configuration.
Krpić, Zdravko; Lukić, Ivica; Köhler, Mirko. Towards a synthetic load generator framework for validating schedules in HPC environment // 2020 19th International Symposium INFOTEH-JAHORINA (INFOTEH), Istočno Sarajevo, Bosna i Hercegovina: IEEE, 2020. pp. 1-6 doi:10.1109/infoteh48170.2020.9066314