1 Comparing Mobility and Predictability of VoIP and WLAN Traces Jeeyoung Kim, Ahmed Helmy Department of CISE, University of Florida, Gainesville, FL, USA {jk2, helmy}@cise.ufl.edu Abstract—How can we obtain realistic mobility models? This has been a question that many researchers have attempted to answer, mostly by analyzing existing WLAN traces. But in the future, will user on-line behavior change with the introduction of new mobile services and devices? We aim to investigate this issue in our study. In this paper, we analyze the mobility of a subset of users different than the general WLAN users; the VoIP users. These users are often mobile while on- line and their devices are always ‘on’. We conjecture that their mobility is captured by the traces better than that of the WLAN users, and we expect them to be highly mobile users representing a trend for future mobile users. To that extent, we contrast the mobility of the VoIP, WLAN users, and three carefully selected sets of users across various metrics. We find that the fraction of time a VoIP user spends at a given AP is lower than WLAN users, indicating that the VoIP users are indeed more mobile than WLAN users. Also, the average and median number of access points visited for VoIP users are 4 to 8 times larger than that of the WLAN users. VoIP users cover a larger area range than WLAN users, indicating that VoIP users are physically more mobile. These findings point to significant difference in mobility characteristics between VoIP users and average WLAN users (commonly used for mobility modeling). In order to examine whether this sharp contrast in mobility affects mobile networking protocols, we compare the performance of different classes of predictors across these different sets of traces. In particular, we evaluate the Markov O(1), O(2), O(3) and the LZ predictors. To our surprise, we find that the average prediction rate is over 60% for general WLAN traces while the prediction success rate drops below 25% for VoIP traces. Lessons learned in our study strongly suggest that both mobility modeling and location prediction should be re-visited in the context of future highly mobile users and devices. Keywords-VoIP; WLAN; mobility; wireless traces; location prediction I. INTRODUCTION Realistic modeling of user mobility is one of the most critical research areas in wireless networks. Mobility data based on real human behaviors may give us the opportunity to improve wireless and mobile services for users in many ways. Currently, several mobility models are proposed based on the analysis of real WLAN traces [1,2,5,6,9]. However, the large collection of WLAN usage traces, seems to capture little mobility from the users. The average user is usually static while using the network, and exhibits a large off time. In this paper, we focus on a subset of the wireless users, who use wireless VoIP devices. These users leave their devices on most of the time and the devices are light enough to carry and use while mobile. Hence, these users show a more mobile characteristic than laptop or other heavy device users while connected to the network. By analyzing these traces we aim to compare behavior of highly mobile VoIP users to the general WLAN users. This sheds light on the realism of WLAN trace-based models. We also aim to examine the effect of any differences on protocol performance, e.g., prediction protocols. Particularly, we compare the mobility of VoIP user traces to whole WLAN traces (as used in previous studies) and also to some samples we have generated based on criteria that distinguish these samples as highly mobile compared to others. We use the metrics of prevalence, number of visited APs and acivity range defined in Section 3 to capture some of the main mobility characteristics of the users in our study. Our results clearly indicate that there is a significant difference between VoIP users and general mobile users, which strongly suggests revisiting mobility models of future always-on portable devices.