Mobility and Predictability of Ultra Mobile Users Jeeyoung Kim and Ahmed Helmy

Mobility Models?  Mobility models: based on WLAN usage traces How realistic are these mobility models? How much mobility do these WLAN traces capture?

Why VoIP?  VoIP Characteristics Devices are on most of the time Devices are light enough to carry around while in use

Ultra mobile?  Are VoIP users good enough? Based on different mobility metrics, we sampled users who are assumed to be ultra mobile (not including VoIP device users) In order to determine the validity of our findings

Data Sets

Data Sets (cont’d)

VoIP vs. WLAN : Mobility Prevalence for VoIP Device UsersPrevalence for General WLAN Users  Prevalence

VoIP vs. WLAN : Mobility # of APs visited for VoIP Device Users# of APs visited for General WLAN Users  Number of APs visited

VoIP vs. WLAN : Mobility Activity Range for VoIP Device UsersActivity Range for General WLAN Users  Activity Range

Predictors: Order-k Markov Assumes location can be predicted from the current context which is the sequence of the k most recent symbols in the location history Markov model represents each state as a context, and transitions represent the possible locations that follow that context. We use Order 1, 2 and 3 predictors for our prediction

Predictors: Lempel-Ziv (LZ) Predicts in the case when the next symbol in the produced sequence is dependent on only its current state (but does not have to correspond to a string of fixed length) Similar to O(k) Markov but k is a variable allowed to grow to infinity

Predictors Each predictor is run for the WLAN movement trace and for each of the ultra mobile test sets including the VoIP trace set The prediction accuracy is measured as the percentage of correct predictions of the next AP to visit

Ultra Mobile vs. WLAN : Predictability Prediction Accuracy Using Markov O(1) for Trace Prediction Accuracy Using Markov O(1) for Trace  Markov O(1)

Ultra Mobile vs. WLAN : Predictability Prediction Accuracy Using Markov O(2) for Trace Prediction Accuracy Using Markov O(2) for Trace  Markov O(2)

Ultra Mobile vs. WLAN : Predictability Prediction Accuracy Using Markov O(3) for Trace Prediction Accuracy Using Markov O(3) for Trace  Markov O(3)

Ultra Mobile vs. WLAN : Predictability  LZ Predictor

Change in Trend VoIP Device Users highly predictable What happened? VoIP users becoming more stationary?  General WLAN Users less predictable More and more light-weight devices are being introduced everyday No longer need to have a VoIP device to use the service

Future Work  Better predictor for “ultra mobile” users 60% is better than 25% but it still is not accurate enough Different flavors of Markov Chain, LZ Family, Prediction using regression methods, etc.  Better mobility models That will incorporate “ultra mobile” users better

Questions?

Thank you!!