首页> 外文会议>IEEE International Conference on Advanced Networks and Telecommuncations Systems >Packet scheduler for meeting deadline in 4-4; 1-4 architecture
【24h】

Packet scheduler for meeting deadline in 4-4; 1-4 architecture

机译:分组调度程序在4-4中会议截止日期; 1-4架构

获取原文

摘要

The distributed computing models such as MapReduce, Dryad, cluster based file systems, query to web search engine, parallel scientific applications and many other are associated with problems such as incast congestion, queue buildup and memory pressure. These adversities hinders the flows to meet their associated deadlines and ultimately the service level agreement (SLA). There are two classes of solutions proposed for the problems of congestion and meeting deadlines for the flows - a congestion control at the transport layer and resource allocation at the network layer. The proposed solutions at the transport layer modulated the TCP window size according to the extent of congestion and the deadline value associated with the flows. These solutions are proposed based on the assumption that the intermediate switches maintain soft-state about the flows. Thus, the entire flow has to follow a static path from begin to end. Contrast to this, the location based routing proposed for DCN tries to utilize the path diversity provided by the underlying architectures by distributing the flows among all the available outgoing paths using the Equal Cost Multiple Paths (ECMP). In this paper, we propose a packet scheduler that helps the flow to meet their associated deadline. We consider our proposed 4-4, 1-4 architecture as an example architecture for location based routing. Packet scheduler for 4-4, 1-4 helps the flows to meet their deadline by arranging the packets at the intermediate switches according to their associated deadlines. With prioritizing and scheduling the packets at the network layer, the flows are not constrained to assign to a static path and the intermediate switches need not to maintain any flow state information.
机译:分布式计算模型,如MapReduce的,树妖,基于集群文件系统,查询网络搜索引擎,并行科学应用及许多其它与诸如聚播拥塞,队列集结和内存压力问题有关。这些逆境阻碍了流动以满足其相关期限和最终的服务水平协议(SLA)。有提议拥堵和会议的最后期限为流动问题的两类解决方案 - 在网络层传输层和资源分配拥塞控制。在传输层提出的解决方案根据拥塞的程度和与该流相关联的最后期限值调制的TCP窗口大小。这些解决方案是基于所述中间交换机维护关于流动软状态的假设提出。因此,整个流程必须遵循从开始到结束的静态路径。与此相反,基于路由的提议DCN位置试图利用由分配由底层架构提供的路径分集使用等价多路径(ECMP)所有可用的出射路径中的流动。在本文中,我们提出了一个数据包调度程序,帮助流动,以满足其相关期限。我们认为,我们所提出的4-4,1-4架构作为一个例子架构基于位置的路由。分组调度器为4-4,1-4有助于流动通过根据它们的相关联的最后期限在中间交换机安排分组,以满足他们的期限。与优先和调度在网络层的数据包,流不限制要分配给一个静态路径和中间交换机不需要维持任何流的状态信息。

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