IMPACT: Investigation of Mobile-user Patterns Across University Campuses using WLAN Trace Analysis

TL;DR

This paper provides a comprehensive analysis of WLAN traces from university campuses, revealing user behavior patterns, encounter dynamics, and network structure insights to improve wireless network modeling.

Contribution

It introduces new metrics and models for understanding individual and group WLAN user behaviors based on extensive campus trace data.

Findings

Most users are active for a small fraction of time

Encounter and friendship relations are highly asymmetric

Encounter patterns follow a biPareto distribution

Abstract

We conduct the most comprehensive study of WLAN traces to date. Measurements collected from four major university campuses are analyzed with the aim of developing fundamental understanding of realistic user behavior in wireless networks. Both individual user and inter-node (group) behaviors are investigated and two classes of metrics are devised to capture the underlying structure of such behaviors. For individual user behavior we observe distinct patterns in which most users are 'on' for a small fraction of the time, the number of access points visited is very small and the overall on-line user mobility is quite low. We clearly identify categories of heavy and light users. In general, users exhibit high degree of similarity over days and weeks. For group behavior, we define metrics for encounter patterns and friendship. Surprisingly, we find that a user, on average, encounters less than 6% of the network user population within a month, and that encounter and friendship relations are highly asymmetric. We establish that number of encounters follows a biPareto distribution, while friendship indexes follow an exponential distribution. We capture the encounter graph using a small world model, the characteristics of which reach steady state after only one day. We hope for our study to have a great impact on realistic modeling of network usage and mobility patterns in wireless networks.