The data set that we use to study mobile apps is signifi-
cantly more diverse geographically and in user base than
previous studies. It covers hundreds of thousands of smartphones
throughout the U.S. in a tier-1 cellular network. This
allows us to make more generalizable conclusions about smartphone
usage patterns.
• We find that a considerable number of popular apps (20%)
are local, in particular, radio and news apps. In terms of traf-
fic volume, these apps are accountable for 2% of the traffic in
the smartphone apps category (i.e., all the marketplace apps
that can be identified by User-Agent) – that is, their user
base is limited to a few U.S. states. This suggests significant
potential for content optimization in such access networks as
LTE and WiFi where content can be placed on servers closer
to clients. Furthermore, it suggests that network operators
need to understand the impact of different app mixes in different
geographical areas to best optimize their network for
user experience.
• Despite this diversity in locality, we also find that there are
similarities across apps in terms of geographic coverage, diurnal
usage patterns, etc. For example, we find that some
apps have a high likelihood of co-occurrence on smartphones
– that is, when a user uses one app, he or she is also likely
to use another one. Users also use several alternatives for the
same type of app (e.g., multiple news apps). These findings
suggest that some apps can be treated as a “bundle” when trying
to optimize for their user experience and that there may
be opportunities for integration.
• We also find that the diurnal patterns of different genres of
apps can be remarkably different. For example, news apps
are much more frequently used in the early morning, sports
apps are more frequently used in the evening, while other
apps have diurnal patterns less visible and their usage is more
flat during a day. These findings suggest that cloud platforms
that host mobile application servers can leverage distinct usage
patterns in classes of apps to maximize the utilization
of their resources. Furthermore, network operators may be
able to leverage these results by optimizing their network for
different apps during different times of the day.
• Mobility patterns can be inferred from network access patterns.
Some apps are more frequently used when users are
moving around; some of them are used more often when
users are stationary. Mobility affects connectivity and performance,
so bandwidth sensitive apps that are mobile may need
to consider techniques to compensate for bandwidth variability.
We find that there is a significant degree of diversity in
the mobility of apps.
The rest of this paper is organized as follows: Related work is
discussed in §2, §3 describes our data set, §4 presents our measurement
results, §5 outlines some implications, and we conclude
our study in §6.