Profiling real PSAs to test scheduling algorithms – running
A significant subset of parallel computations are PSAs (Parameter Sweep Application), where a calculation has to be performed for various values of a parameter vector independently, without inter-communication.
The running time of the calculations for different parameters may vary considerably, so scheduling optimally a PSA on parallel machines in general is intractable.
There is a rich literature in operation research discussing scheduling approaches.
Our main goal of research is designing new scheduling algorithms adapted for executing PSAs in parallel infrastructures.
The algorithm will be tested on both synthetic and on real-world data.
The goal of this project is to collect detailed runtime profile of real, previously published PSAs, allowing the testing of our algorithms with real dataset.
To this end, we re-run modified PSA programs which, in addition to performing calculations, provide detailed information on the running time.
- Project owner:
- Kápolnai Richárd Péter (Irányítástechnika és Informatika Tanszék)
- Irányítástechnika és Informatika Tanszék (VIK-IIT)