Congestion-aware task mapping in NoCbased MPSoCs with dynamic workload

Applications running in heterogeneous MPSoCs, as mul-timedia and networking, normally contain a dynamic work-load of tasks. This implies a varying number of tasks simul-taneously running, with their number possibly exceeding the available resources. This may require the execution of task mapping at run time, to meet real-time constraints. Most works in the literature propose the use of static mapping [1][2], where the best placement of tasks is defined at design time. Consequently, such methods are not appropri-ate for dynamical workloads. Task migration [3][4] has also been used in heterogeneous MPSoCs to optimize the per-formance at run-time. It consists either in relocating tasks when a performance bottleneck is detected, or to distribute the workload more homogeneously among the MPSoC proc-essors. Differently from task migration, dynamic mapping can insert new tasks into the system at run time. This work investigates the performance of different map-ping algorithms in NoC-based MPSoCs with dynamic work-load. The main cost function in mapping algorithms is to optimize the occupation of the NoC links. It is possible to achieve performance gains if the mapping algorithm is able to minimize NoC congestion. Without loss of generality, heterogeneous MPSoC archi-tectures may be represented, as a set of processing nodes that interact via a communication network. Processing nodes may support either hardware or software task execution. Hardware tasks execute in reconfigurable logic (reconfigurable areas) or dedicated IPs. If reconfigurable logic is used, the hardware presents flexibility similar to software. It becomes possible to load the hardware tasks on-the-fly using dynamic reconfigu-ration. Software tasks execute in instruction set processor, ISPs. The number of tasks may exceed the MPSoC resources