Master-Seminar: Performance Modeling for Parallel Applications and Emerging Architectures (IN2107)
Prof. Dr. Martin Schulz, Urvij Saroliya, Yi Ju, Muhammad Arslan Ansari, Felix Jung, Stepan Vanecek
| Pre-course meeting: | Date: 05.02.2025 Time: 11:00 Location: tum-conf.zoom-x.de/j/69731968659 |
| Kick-off meeting: | tbd |
| ECTS: | 5 |
| Language: | English |
| Type: | Seminar, 2S |
| Moodle course: | tbd |
| Registration: | Matching System |
Performance models are utilized by professionals from diverse fields. One of their most common applications is by software developers, who use them to estimate the upper limits of performance. With this insight, developers can evaluate the quality of their code and its efficiency in utilizing available hardware resources. Consequently, performance models aid in identifying bottlenecks and directing efforts to optimize application performance. Beyond this, performance models are also employed in processor design, where they facilitate exploration of the design space to identify promising configurations.
In this seminar, we would like to review existing Performance Models and identify following aspects:
(i) benefits of using a certain model for its interpretability, accuracy, and portability
(ii) major pitfalls and range of applicability
(iii) potential directions for improvements
Potential Topics
- Analytical machine models
- Machine Learning-based modeling techniques
- Architecture-specific techniques (GPU / FPGA)
- Application Domain-specific techniques (Cloud, HPC)
- Modeling for (MPI-based) Large-scale Applications
- Energy / Carbon-aware Models
- Performance Analysis for Quantum Computing Software Stack