首页> 外文期刊>Concurrency and Computation >Providing differentiated services, congestionmanagement, and deadlock freedom in dragonfly networks with adaptive routing
【24h】

Providing differentiated services, congestionmanagement, and deadlock freedom in dragonfly networks with adaptive routing

机译:通过自适应路由在蜻蜓网络中提供差异化​​服务,拥塞管理和死锁自由

获取原文
获取原文并翻译 | 示例

摘要

The number of endnodes in high-performance computing systems has grown significantly in thernlast years. Hence, the interconnection network has become an essential issue as it may end uprnbeing the system bottleneck if it is not properly designed. In that sense, the Dragonfly topologyrnhas become very popular for interconnecting high-performance computing systems in the lastrnyearsbecause it offers high performance at an affordable cost.However,whenusing deterministicrnminimal-path routing, this topology is not able to offer a high performance under certain trafficrnconditions. This problem can be solved by using oblivious or adaptive routing. However, therernare no congestionmanagement techniques specially tailored toDragonfly topologies using obliviousrnor adaptive routing. Note that in congestion situations, the Dragonfly performance may droprnbecause of the head-of-line blocking effect. This effect could be evenmore dangerous in systemsrnwhere several applications with different priorities coexist. In thiswork we propose several techniquesrnespecially designed for providing differentiated services and congestion management inrnDragonfly networks using oblivious or adaptive routing. First, we propose the hierarchical 3-levelrnqueuing queuing scheme, which configures several virtual channels distributed into 3 virtual networksrnto reduce the head-of-line blocking while deadlocks derived from the routing algorithmrnare prevented. Second, we extend hierarchical 3-level queuing to provide differentiated servicesrnthrough 2 different solutions. Finally, some experiments are performed to show the benefitsrnobtained by using the proposed techniques.
机译:在最近几年中,高性能计算系统中的端节点数量显着增长。因此,互连网络已成为必不可少的问题,因为如果设计不当,互连网络可能最终会成为系统瓶颈。从这个意义上讲,Dragonfly拓扑在过去的几年中已成为用于互连高性能计算系统的一种非常流行的方法,因为它以可承受的价格提供了高性能。 。通过使用遗忘或自适应路由可以解决此问题。但是,没有专门使用蜻蜓或自适应路由为蜻蜓拓扑量身定制的拥塞管理技术。请注意,在拥挤的情况下,蜻蜓的性能可能会由于行首阻塞效应而下降。在多个具有不同优先级的应用程序共存的系统中,这种影响可能更加危险。在这项工作中,我们提出了几种技术,这些技术是专为使用遗忘或自适应路由在蜻蜓网络中提供差异化​​服务和拥塞管理而设计的。首先,我们提出了一种分层的三级排队队列方案,该方案配置了分布到3个虚拟网络中的多个虚拟通道,以减少行头阻塞,同时避免了从路由算法派生的死锁。其次,我们通过3种不同的解决方案扩展了分级3级排队以提供差异化​​服务。最后,进行了一些实验以显示通过使用所提出的技术获得的益处。

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