基于拥塞预测的NoC自适应仲裁方法*

传统用于总线系统或互联网的仲裁方法已不能很好地适应NoC应用环境。围绕NoC系统性能的关键影响因素——拥塞状态，提出了一种基于全局和本地拥塞预测的仲裁策略(GLCA)，以改善NoC网络延迟。实验结果表明，相对于RR方法，新仲裁算法使得网络平均包延迟和平均吞吐量最大分别可改善20.5%和8%，并且在不同负载条件下都保持了其优势。综合结果显示， GLCA与RR方法相比，路由器仅在组合逻辑上有少许增加(25.7%)。

NoC adaptive arbitrating policy based on congestion prediction

Many conventional arbitration policies used in bus-based systems or macro network were not suitable for NoC because of its new features. It was found that congestion was a common phenomenon in NoC communication and posed great impact on the network delay of NoC. This paper proposed a new arbitration policy-GLCA based on global and local congestion prediction to improve the network latency. Experiment results demonstrat that the new policy fits well with NoC. Compared with round robin arbitration policy, average latency per packet and average throughput can be improved by 20.5% and 8% at best, respectively. GLCA always keeps its advantage under different load conditions. In synthesis, the cost of router with GLCA is only larger than that with RR by 25.7% in combinational logic.