NoC中基于路径多样性的交叉开关细粒度容错设计

发布时间：2018-01-12 14:41

本文关键词：NoC中基于路径多样性的交叉开关细粒度容错设计　出处：《计算机辅助设计与图形学学报》2017年01期 　论文类型：期刊论文

【摘要】：在片上网络中,交叉开关是路由器的重要组成部分,当其出现故障会导致数据无法传到下一跳,易造成资源浪费,影响整个网络的性能.对此,通过详细考虑交叉开关中数据分配器、内部链路以及数据选择器的故障情况,充分利用交叉开关的可用资源,提出交叉开关细粒度容错设计方案,实现容错路径的多样性.当数据到达数据分配器时,查看故障信息表,若存在故障则顺序选择一个无故障的数据分配器传输数据;当数据经过数据分配器传至内部链路时,若存在链路故障,同样查找故障信息表快速选择合适的容错链路进行传输;最后数据到达数据选择器,若此时数据选择器出现故障,数据则经过额外的数据选择器到达输出端口.实验结果表明,故障率逐渐增大的时候,该方案比已有方法平均延时降低了6%~36%,吞吐率提高了6.7%~27.7%,且容错效果不会随着故障率增大而显著下降.
[Abstract]:In the on-chip network, crossover switch is an important part of the router. When it fails, the data can not be transmitted to the next jump, which can easily lead to the waste of resources and affect the performance of the whole network. By considering the fault of the data distributor, internal link and data selector in the cross switch in detail, the fine grain fault tolerant design scheme of the cross switch is put forward by making full use of the available resources of the cross switch. When the data arrives at the data allocator, the fault information table is checked, and if there is a fault, a faultless data allocator is selected in order to transmit the data. When the data is transmitted to the internal link through the data allocator, if there is a link failure, the fault information table is also found to quickly select the appropriate fault-tolerant link for transmission. Finally, the data reaches the data selector. If the data selector fails, the data will reach the output port through the additional data selector. The experimental results show that the failure rate increases gradually. Compared with the existing methods, the average delay of this scheme is reduced by 6 / 36, and the throughput is increased by 6.7and 27.7. moreover, the fault-tolerant effect does not decrease significantly with the increase of the failure rate.
【作者单位】：合肥工业大学计算机与信息学院;合肥工业大学电子科学与应用物理学院;
【基金】：国家自然科学基金(61474036,61274036,61371025,61574052) 安徽省自然科学基金(1508085MF117)
【分类号】：TN47
【正文快照】： 随着集成电路技术的发展,单个芯片上会出现由数十甚至数百个处理器联合工作的超复杂电路系统.然而基于共享总线机制的传统片上系统(system-on-chip,So C)的体系结构无法解决复杂多样的系统通信需求,如互连线延时、系统全局同步和设计效率.为解决So C的瓶颈问题,一些研究人员借

【相似文献】