Title: Passive Measurement Studies of Large-scale IP Networks

Tongqing Qiu

Committee:

Dr. Jun Xu, College of Computing, Georgia Institute of Technology (Advisor)
Dr. Mostafa H. Ammar, College of Computing, Georgia Institute of Technology
Dr. Nick Feamster, College of Computing, Georgia Institute of Technology
Dr. Xiaoli Ma, School of Electrical and Computer Engineering, Georgia Institute of Technology
Dr. Jia Wang, AT&T-Lab Research

Abstract:

Large-scale IP networks (e.g. backbone network, commercial IPTV network) are designed with the goal of providing high availability and low delay/loss while keeping operational complexity and cost low. Meeting this goal requires network operators to perform a wide range of measurement studies to understand network status and dynamics. Although a large number of passive and active measurement tools and techniques are used, the problem of building a comprehensive and integrated monitoring infrastructure to address all ISP’s needs (ranging from performance monitoring, anomaly diagnosis to network planning) is far from solved.

In this dissertation, we focus on the passive measurement study using the following general method. We collect the measurement data from distributed network elements, and apply advanced statistical methods tailored to network domain to uncover meaningful information from the data set. Our three studies fulfill three requirements of network measurement respectively: performance monitoring, troubleshooting, and network design and planning. First, we propose a novel methodology to infer the delay distribution passively without any perturbation to real traffic. Second, we design a system that can automatically transform and compress low-level syslog messages into meaningful prioritized network events. It can provide critical input to network troubleshooting and visualization. Finally, we analyze and model user activities in an operational nation-wide IPTV network, and design a workload generator which takes a small number of input parameters and generates synthetic trace that mimic aggregated user behavior. The generator can estimate the unicast and multicast traffic accurately, proving itself as a useful tool in driving network design and planning study. We believe that our measurement studies make a solid step towards building an ideal passive measurement infrastructure.