Figure 2: CPU utilization and energy consumption
during one scrolling operation on a webpage.
endeavored to find an energy-efficient strategy to dynami-
cally adjusting the CPU frequency and voltage based on the
concept of DVFS[6][23].
Radio Power Consumption: since communication is
the basic function of smartphones, the power consumption
of wireless module draws a lot attention in researching area.
Niranjan et al. first found the tail[4] overhead which is
caused by the lingering in high power states after complet-
ing a transfer. Then, in 2012, Qian et al. elaborately sum-
marized several optimizing schemes[22][21] on tail energy.
Besides, Athivarapu et al. explored a method to reduce the
radio usage by monitoring the program execution pattern[3].
In another direction, Schulman et al. discussed the impact
of signal strength on communication energy in [25].
Application Power Consumption: with the Explosive
growth of smartphone applications, increasingly researchers
are interested in investigating the energy consumption in ap-
plications. Narseo made a statistical analysis on the energy
consumption of different types of Apps in [26]. Besides, [13]
discussed the impact on energy consumption of cooperation
between Apps and operating system. [18][20] proposed the
concept of energy bug in smartphones and gave a first trial
on diagnosing energy bugs.
Display Power Consumption: due to the physical char-
acteristics of display hardware, energy saving could be achi-
eved by wisely adjusting LCD and OLED display param-
eters. Existing work studies screen hardware power con-
sumption through studying the screen power model[16], cal-
ibrating the backlight level[2] or the display color scheme[9],
while our work explores the power consumption of mobile
device made by exorbitant screen frame rate.
Although the battery-life issues of mobile devices have
gained much attention, the energy cost made by human-
screen interactions such as scrolling remains elusive. There
are some studies on frame rate recently, but these works
mainly focus on how to improve frame rate through hard-
ware and software[28] or how to use frame rate as a QoS
metric[8][15]. Therefore, existing work has not concerned
the impact of frame rate to the energy consumption.
3. EMPIRICAL STUDIES & TRACE ANAL-
YSIS
We first conduct an empirical study to show that human-
screen interactions may be a new element in the spectrum of
power consumption on smartphones. Among all normal in-
Figure 3: Scrolling time ratio and energy consump-
tion ratio of several popular smartphone applica-
tions.
teraction operations, scrolling is a very typical one to display
the content of interest on screen. Moreover, the interaction-
intensive applications such as browsing and reading are more
likely to have a higher ratio of scrolling to click (1.4 in Angry-
Birds and 1.1 in CNN)[27]. Besides, the energy consumption
caused by click is much lower than scrolling. In this section,
we explore the influence of scrolling operations to the energy
consumption on smartphones.
3.1 Power Consumption Caused by Scrolling
on Screen
To illustrate the relationship between screen scrolling op-
erations and power consumption, we first examine one scro-
lling operation when browsing a webpage using a typical
smartphone (Nexus S). During this operation, we record
the CPU utilization and the measurement of energy con-
sumption then plot the results in Figure 2. It is clear to
see that, once the scrolling starts, the CPU utilization im-
mediately increases to 100%. Meanwhile, the power jumps
up twice higher than usual. Moreover, the CPU utilization
and power consumption keep high until the scrolling action
ends. Similar results are obtained through more trials of this
experiment with different users and mobile devices. Fur-
thermore, we also conduct this experiment with different
scrolling speeds and directions, and get the same results.
We then further examine the impact of scrolling opera-
tions among different most popular applications on different
smartphones. Specially, we randomly select 327 volunteers
(i.e., 56 faculty members and 271 students) on campus dur-
ing lunch time, and let them try at least five applications in-