路由器
仲裁
计算机网络
计算机科学
路径(计算)
网络拥塞
网络数据包
延迟(音频)
芯片上的网络
吞吐量
电信
无线
政治学
法学
作者
Lu Liu,Yadong Sun,Zhangming Zhu,Yintang Yang
标识
DOI:10.1088/1674-4926/39/12/125006
摘要
To meet the demand for high on-chip network performance, flexible routing algorithms supplying path diversity and congestion alleviation are required. We propose a CAOE-FA router as a combination of congestion-awareness and fair arbitration. Buffer occupancies from downstream neighbors are collected to indicate the congestion levels, among the candidate outputs permitted by the odd-even (OE) turn model, the lightest loaded direction is selected; fair arbitration is employed for the condition of the same congestion level to replace random selection. Experimental results show that the CAOE-FA can reduce the average packet latency by up to 22.18% and improve the network throughput by up to 68.58%, with ignorable price of hardware cost.
科研通智能强力驱动
Strongly Powered by AbleSci AI