Bozhidar Kolarov: Multiobjective task mapping in a multi-core architecture (Diplomarbeit)

05.07.2022, 14:00 Uhr

Einladung zur Präsentation von Herrn Bozhidar Kolarov

Thema: Multiobjective task mapping in a multi-core architecture

Projekt: Diplomarbeit

Betreuer: Gökhan Akgün, Cornelia Wulf, Michael Willig

Abstract: While running applications with high performance, power dissipation can be dynamically reduced with scaling methods. At the same time, the communication latency in a multi-core system should not be ignored. A mapping strategy can be used to establish an optimal operating state on reconfigurable systems. In this thesis, a NoC-based multi-core system was designed and a mapping strategy for for this design was implemented and evaluated. In the mapping strategy, a master distributes tasks to processing elements (PEs) for further execution. Besides the mapping strategy, it should take into account the order in which the tasks are executed in PEs and also meet real-time requirements. The dynamic change of frequency affects the execution time of tasks. Therefore, the effect has to be supported by the strategy as well. In addition, there is inter-task communication between PEs in the NoC system. The resulting latency is also considered by the mapping algorithm. The architecture can be used for emulating an FPGA cluster. Thereby the communication latencies are annotated at the edges. The results are evaluated with the BEEBS benchmark.