Locating Causal Segments Along A Performance-Degraded End-to-End Path

Locating Causal Segments Along A Performance-Degraded End-to-End Path
复制标题

DOI:
--
复制
发表时间:
2009
期刊:
--
影响因子:
--
通讯作者:
Yuki Sakai;M. Uchida;M. Tsuru;Y. Oie
Yuki Sakai;M. Uchida;M. Tsuru;Y. Oie
中科院分区:
其他
文献类型:
--
作者:
Yuki Sakai;M. Uchida;M. Tsuru;Y. Oie

文献摘要

相似文献

由于网络资源的统计和松散控制份额的固有特征,由分组延迟变化和/或分组丢失(通常称为“拥塞”)导致的瞬时网络性能(质量)劣化仍然可能发生在因特网上。通常,端到端路径的性能不对单个Internet服务提供商(ISP)负责,因为该路径在两端穿过多个ISP和用户管理的网络。因此,对于端到端服务应用用户和/或提供商来说,自行找到瞬时网络性能下降的原因具有实际重要性。针对这一问题,本文提出并实现了一种基于网络断层扫描的实用工具。该工具运行在性能下降的端到端路径两侧的用户主机上,并推断每个方向段上的丢包率和延迟方差,以最终定位路径上的因果段沿着。这里的段定义为由两个路由器分割的网络部分,由任意数量的路由器、交换机和/或物理链路组成。我们的工具可能看起来像pathping,一个基于Windows的网络命令,结合了traceroute和ping [1]。然而,我们的工具可以更定量和精确地推断每个部分的性能状态,使用理论基础与单向和往返测量的组合。
Transient network performance (quality) degradation resulting from packet delay variations and/or packet losses, often called as “congestion”, is still likely to occur on the Internet due to the inherent feature of statistical and loosely controlled shares of network resources. Typically, the performance of an end-to-end path is not responsible to a single Internet Service Provider (ISP) because the path traverses multiple ISPs and user-managed networks at both sides. Thus, it is of practical importance for end-to-end service application users and/or providers to find the causes of transient network performance degradation by themselves. In response to this problem, in this paper, we propose and implement a practical tool based on network tomography. This tool runs on user hosts at both sides of a performancedegraded end-to-end path and infers the packet loss rate and delay variance on each individual directional segment to finally locate the causal segments along the path. A segment here is defined as a network portion segmented by two routers and consisting of a sequence of any number of routers, switches, and/or physical links. Our tool might look like pathping, an Windows-based network command that combines traceroute and ping [1]. However, our tool can more quantitatively and precisely infer the performance status of each segment using a theoretical basis with combination of one-way and round-trip measurements.