首页> 外文会议>International Conference on Knowledge-Based Engineering and Innovation >A fault-tolerant routing algorithm in 3D topology manycore processors
【24h】

A fault-tolerant routing algorithm in 3D topology manycore processors

机译:3D拓扑多核处理器中的容错路由算法

获取原文

摘要

The unprecedented progress in semiconductor technology has provided great opportunities for commercialized computationally intensive applications. Amdahl's law was applied for multiprocessor computers till several years ago but his laws are now useful to help us understand and develop using manycore chip multiprocessors (CMP). Obviously manycore-based designs could not be done blindfold and it needs detailed calculations. In CMP's with hundred processing cores, 3D topology in the form of network-on-chip (NoC) can be used for shortening the wires length leads to low latency, low power dissipation and scalability. Meanwhile faults can occur in NoC both at the router and in communicational links. There are many fault-tolerant solutions that their function is based on rerouting the packets. In this paper we propose a fault-tolerant technique which is completely adaptive and use available non-broken links. The focus of this technique is keeping the performance of NoC when there is a faulty link and the packets from a source to a destination never get lost. Experimental results shows that this algorithm can tolerate more than 10 faulty links in different parts of NoC and it can achieve more than 97% reliability.
机译:半导体技术的空前进步为商业化计算密集型应用提供了巨大的机会。直到几年前,阿姆达尔定律一直应用于多处理器计算机,但是现在他的定律对于帮助我们理解和开发使用多核芯片多处理器(CMP)很有用。显然,许多基于核的设计无法被蒙住眼睛,因此需要详细的计算。在具有数百个处理核心的CMP中,可以使用片上网络(NoC)形式的3D拓扑来缩短导线长度,从而导致低延迟,低功耗和可扩展性。同时,路由器和通信链路中的NoC都可能发生故障。有许多容错解决方案,其功能基于重新路由数据包。在本文中,我们提出了一种完全自适应的容错技术,并使用可用的非中断链接。该技术的重点是当链路出现故障并且从源到目的地的数据包永不丢失时,保持NoC的性能。实验结果表明,该算法在NoC的不同部分可以容忍10条以上的故障链路,并且可以达到97%以上的可靠性。

著录项

相似文献

  • 外文文献
  • 中文文献
  • 专利
获取原文

客服邮箱:kefu@zhangqiaokeyan.com

京公网安备:11010802029741号 ICP备案号:京ICP备15016152号-6 六维联合信息科技 (北京) 有限公司©版权所有
  • 客服微信

  • 服务号