Mobile device-level data modeling through high utilization mobile applications

Capacity level modeling of mobile device bandwidth requirements employing high utilization mobile applications

“…We observe that the Folded Normal distribution, which was able to capture the average data amounts in [15] fairly well, now results in significantly higher simulated data amounts while the variability remains similar. We conclude that the different identification approach of high utilization applications has rendered this prior distribution no longer suitable.…”

“…We observe that only 12% of the total network traffic originates from these high utilization applications and that their overall number of active bins varies significantly. As reported in [15], utilizing this clustering outcome we were able to achieve a close fit overall for average amounts of data, but lacked a sufficient estimation of the network traffic variability on these longer time scales. In turn, we are motivated to modify the clustering basis to derive a more evenly distributed representation.…”

“…Typical applications in the highest data utilization class include, e.g., the Dropbox client 'com.dropbox.android'. We originally reported on this clustering outcome in [15].…”

“…We note that while we identify the four states for the background applications as well, we do not need to derive the associated state transition probability matrix. Our model assumes that the background application aggregate state results from the highly active application aggregate state, which is an extension of the model we described in [15]. The determination of the states in both categories allows to compare their occurrences in the submitted user data and determine the state-dependent average and standard devia- , respectively.…”