Summary
Asia-Pacific Network Operations and Management Symposium
2019
Session Number:TS2
Session:
Number:TS2-3
Locating Delay Fluctuation-Prone Links by Packet Arrival Intervals in OpenFlow Networks
Nguyen Minh Tri, Syunya Nagata, Masato Tsuru,
pp.-
Publication Date:2019/9/18
Online ISSN:2188-5079
Summary:
In cloud computing and content delivery networking, OpenFlow-based centrally managed networks to connect distributed servers are becoming popular these days. To maintain service quality and availability in such networks by flexible and dynamic traffic engineering, detecting and locating deteriorated (e.g., congested) links in an efficient manner is essential. Following our previous study that actively monitors packet loss rate to find deteriorated links, in this paper, we actively estimate packet delay variance on each link (note both up and down directions of each full-duplex link are distinguished) in an OpenFlow network. A notable feature is that packet delay variance is estimated based on monitoring arrival time intervals of probe packets without directly measuring packet delay time over a link. In the proposed scheme, a series of probe packets is launched from a measurement host and traverses each direction of each link once and only once by multicasting, while arrival time intervals of those packets at each input port of OpenFlow switches are monitored. Then the OpenFlow controller collects the arrival time interval statistics from those switches to locate delay fluctuation-prone links, i.e., links with a high packet delay variance, which are likely congested or physically unstable. In addition, to minimize the necessary number of accesses to switch ports, an appropriate order of collecting statistics from switches is dynamically controlled. The results of numerical simulation on large-scale network topologies demonstrate the effectiveness of our proposed scheme. A prototype implementation which requires an extension of OpenFlow is also presented on Mininet.