RASMAP: An efficient heuristic application mapping algorithm for network-on-chips

Abstract

This paper proposes and evaluates a performance efficient application mapping algorithm for mesh-based NoCs. The proposed algorithm first prioritizes tasks of the given application graph based on their total in/out communication traffic. Then a task with the most communication traffic is selected and mapped onto the center part of the mesh topology i.e., a core with the most available communication channels. After that, repeatedly, the next task is selected in a way that has the most communications with the already mapped tasks. Such a task is mapped onto the core which its degree is proportional to the tasks link degree. The proposed method is evaluated by Noxim which is a cycle-accurate NoC simulator in terms of communication cost i.e., to total number of packets traversed through the network to complete the application graph. The proposed method is compared with several previously proposed mapping algorithms including NMAP, CMAP, LMAP, PSMAP, and CASTNET. Comparisons show that the proposed method offers better performance and consumes lower energy in the network.