高级检索

基于故障链路缓存再利用的NoC容错路由算法

Fault-tolerant Routing Algorithm of NoC Based on Buffer Reuse of Faulty Links

  • 摘要: 建立故障模型是进行片上网络容错研究的基础,传统的细粒度故障模型未能有效地区分链路故障和通道故障.为了进一步提高片上资源的利用率,构建了一种粒度更细的微粒度故障模型,并在该模型的基础上提出了基于故障链路缓存再利用的容错路由算法.该算法为每个通信节点增加4条自收发通道,并采用基于缓存再利用的透传机制,通过复用故障链路两端的正常缓存和通道来透传故障通道上的数据包,提高了数据包采用最优输出端口的概率.实验结果表明,文中算法在高故障比例的片上网络中优势明显,且能以相对较小的硬件开销换取平均吞吐量、平均延迟和数据包平均跳数等性能的大幅度提升.

     

    Abstract: Fault model is the basis of fault-tolerant research on network on chip (NoC) .Since the conventional fine-grained fault model could not distinguish link failures from channel failures effectively, a finer grained fault model, which is named very fine-grained fault model, is established in this paper to further improve the utilization rate of NoC resources, and a fault-tolerant routing algorithm based on buffer reuse of faulty links (FRA-BRFL) is also proposed.The algorithm adds 4 self-communicating channels (SCC) for each router, and adopts transparent transmission mechanism based on buffer reuse (TTBR) .Reusing normal buffers and channels at both ends of the faulty link to transmit packets of faulty channels transparently, TTBR can increase the probability of packets outputting with optimal port.The experimental results demonstrate that, FRA-BRFL has obvious advantages in NoC with high failure rate, and the algorithm can perform better in average throughput, average latency and average hop counts with lower hardware overheads.

     

/

返回文章
返回