Algoritmos de particionamento para MPSoCs heterogêneos baseados em NoC

Abstract

Several new applications are composed by heterogeneity of tasks implying high complexity degree, and requiring high processing and communicating rate. Multiprocessor System-on-Chip (MPSoC) based on Network-on-Chip (NoC) is a promising targeting architecture to fulfill these requirements, due to its high computation and communication parallelism that enables several tasks executed at the same time. Furthermore, these applications requirements are better fulfilled by MPSoC composed by different types of processors heterogeneous MPSoC. One challenge in current heterogeneous MPSoC design is partitioning of application tasks, aiming energy consumption minimization and fair load balance. This work contribution is twofold: (i) analysis and comparison of partitioning algorithms; and (ii) the evaluation of partitioning as a pre-mapping task. This work analyzes and compares stochastic and new heuristic partitioning algorithms for obtaining low energy consumption and efficient load balance when applied to tasks partitioning onto heterogeneous MPSoC. In addition, performance results obtained from simulations indicate that the static partitioning technique can be used on application tasks before mapping activities to improve the quality on the static or dynamic mapping and also for minimizing processing time.