Run-time mapping of multiple communicating tasks on MPSoC platforms

Abstract

Multi-task supported processing elements (PEs) are required in a Multiprocessor System-on-Chip platform for better scalability, power consumption etc. Efficient utilization of PEs needs intelligent mapping of tasks onto them. The job becomes more challenging when the workload of tasks is dynamic. These scenarios require tasks to be mapped at run-time. This paper presents a run-time mapping technique for efficiently mapping the tasks of applications on the multitasking resources. The technique tries to map the communicating tasks onto the same processing resource and also the tasks of an application close to each other in order to reduce the communication overhead. For an evaluated scenario, the presented technique reduces total execution time by 22%, average channel load by 47% and power dissipation by 48% when compared to state-of-the-art run-time mapping techniques.