Adaptive routing using context-aware agents for networks on chips

In this paper we propose an adaptive routing algorithm which is based on deterministic XY routing algorithm. In our model a switch is a context-aware agent and a network is a society of context-aware agents which are ever learning and adapting to distribute the congestion uniformly and isolate the malformed switches (agents). In conventional XY routing, first, the load in the center of a network is much higher rather than total average and this leads to hot spot in the center of network. And second, a malfunction in switches could make part of network out of access. But in our proposed routing, first, all agents are aware from their neighbor's congestion and collaboratively try to route their input packets through less congested route, according to their experiences learned before and second, when a new malfunction takes place in the network, all agents collaborate each other to recognize the malformed agent and learn the best route. The main objective of our routing algorithm is to distribute network load uniformly throughout the network. We developed a NoC environment using SystemC and applied two multimedia applications and a random traffic to proposed routing and compared with classic XY routing. Experimental results show better latency while average load is unchanged.