Scheduling Data Communication based Services
on the Personal Mobile Devices
Setia Budi
1
and Vishv Malhotra
2
1
Faculty of Information Technology, Maranatha Christian University, Bandung, Indonesia
2
School of Computing and Information System, University of Tasmania, Hobart, Australia
Keywords: Scheduling, Multi-objective Optimization, Mobile Device, Real-time Data Access.
Abstract: Mobile devices have become more and more popular, and the services have grown in number and range.
Ready access to the Internet is one of the characteristics of the mobile devices that deliver significant value
for their users. However, the users are also concerned about costs and other factors related to this access.
This paper describes a multi-objective model to optimally schedule a service on a mobile device that
requires data communication with the external data repositories. These objectives are incomparable with
each other and represent different personal and temporal needs and preferences. Thus, the objectives cannot
be assigned unanimously accepted fixed weights to generate a single outcome metric. Pareto optimal
solutions are used to provide the best option for determining efficient schedules.
1 INTRODUCTION
Mobile device usage is growing as is their ability to
provide sophisticated services. The primary reason
for the growing popularity of the mobile devices is
their ability to support relationships among the peers
through the connectivity and communication
anywhere and anytime mantra (Constantiou, 2007;
Cui and Roto, 2008; Sarker and Wells, 2003).
However, the mobile devices have noticeable
resource constraints and users are often concerned
about the communication costs, battery life and
device busy periods. The paper describes a model
that aids in finding Pareto optimal schedules for the
data communication dependent services available
through the mobile devices.
To explain the reader about the scheduling
problem, we present the following made-up story: A
mobile device user accidently meets the President
holidaying on a remote island and is invited for a
breakfast with the first family. The user has a
number of pictures that she is eager to share them
with her social groups. What is the best schedule to
upload these pictures to match the urgency to upload
tempered with limited access to the bandwidth and
perhaps desire to conserve the device battery?
Though the story is a bit unrealistic, everyday
mobile device users face the similar sentiments
when using their devices.
The rest of the paper is organized as follows.
Section 2 presents the Multi-objective Model for
Mobile Data Communication. Each objective
functions including related constraints are presented
in Section 2.1 to Section 2.5. In Section 3, a brief
description of the experimental evaluation for our
model is presented. We conclude the paper in
Section 4.
2 MULTI-OBJECTIVE MODEL
FOR MOBILE DATA
COMMUNICATION
Several studies have been conducted to identify and
understand the satisfaction factors of the mobile
device users (Büyüközkan, 2009; Cui and Roto,
2008; Sarker and Wells, 2003). However, these
studies do not explain how the mobile device users
can optimize the access to a service by scheduling it
for better outcomes. The problem of optimizing the
access to a service through a mobile device can be
set as a multi-objective optimization problem since
there are more than one satisfaction factors required
to be optimized and these satisfaction factors are
conflicting with each other.
Two primary decision variables controlling the
access to mobile services are: the postpone interval
401
Budi S. and Malhotra V..
Scheduling Data Communication based Services on the Personal Mobile Devices.
DOI: 10.5220/0004412004010408
In Proceedings of the 15th International Conference on Enterprise Information Systems (ICEIS-2013), pages 401-408
ISBN: 978-989-8565-59-4
Copyright
c
2013 SCITEPRESS (Science and Technology Publications, Lda.)
(delay) and the communication mode to use for the
data communication (mode). The postpone interval,
delay, represents the length of the period by which
the data communication is delayed after the need for
the data communication outside the mobile device is
first noted; in this paper, the time at which the data
communication need is first noted is marked as time
0. The communication mode, mode, as the second
decision variable represents the use of a wired, Wi-
Fi or a cellular data network to contact the data
repository. In this paper, we will use the following
identifiers and functions in the equations:
size(data) Size of the data to be
communicated by the
mobile service.
speed(mode) Communication speed
for communication
mode mode.
batteryCharge(delay) MilliAmpere hours
(mAh) of the electric
charge stored in the
mobile device battery at
time delay.
dischargeRate(mode) Battery discharge rate
mAh/s in
communication mode
mode.
qdr Quiescent discharge rate
for the device battery to
keep the device active.
nextRechargeTime(delay) Time to the next battery
recharge opportunity
after delay interval
delay.
accessTime(delay) Time to start
transmitting the data at
or after time delay.
schedule(mode) Availability schedule
to utilize
communication mode
mode.
available(delay,mode) Boolean function; true if
the mobile user can use
communication mode
mode at delay interval
delay. A mobile user
may avoid a live
communication mode if
it is considered
unsecure. Thus, the
function captures
security needs.
fixedCosts(delay,mode) Fixed costs to utilize the
communication mode
mode at delay interval
delay.
varCosts(delay,mode) Variable costs to utilize
the communication
mode mode at delay
interval delay.
When accessing a service requiring data
communication from a mobile device, the device
owner seeks to choose values for the decision
variables delay
and mode to use the best data
communication outcomes for the service. The
proposed model defines five objectives to find the
Pareto-optimal solutions: data access cost, service
efficacy, battery charge depletion, depleted battery
charge encumbrance, and device unavailable
duration.
2.1 Minimizing Data Access Cost
A high cost for the mobile Internet access
discourages the users from accessing the Internet
from their mobile device (Büyüközkan, 2009;
Blechar, 2006). The cost of the mobile Internet
access is determined by the amount of data
communicated and is dependent on the available
communication modes and the time of the access.
Since the cost of communication varies across many
different network providers, a lower data
communication cost can be achieved by delaying the
communication for a particular interval of time in
order to get a cheaper available connection. In
mathematical notation the cost objective goal can be
represented as:
,
where
,
,
,
∗
(1)
2.2 Maximizing Service Efficacy
Mobile users and applications place a significant
ICEIS2013-15thInternationalConferenceonEnterpriseInformationSystems
402
premium on the immediacy of the data interaction
and servicing. In this respect, the data
communication requirements have the
characteristics similar to the real-time applications.
The full benefits of the service are derived if the data
can be transferred before a user or service specific
deadline; titled festive deadline. However, if the
service is delayed past a later deadline, titled
obsolesce deadline, the mobile user derives little
benefit from the service. Basic real-time schedulers
approximate the benefits from a service completed
between these two deadlines using a linearly
decaying service efficacy function (Laplante, 2004;
Williams, 2006). Thus, the service efficacy objective
is expressed as:
,
where
,
100,
,
0,
(2)
2.3 Minimize Battery Charge Depletion
Battery status is a fundamental constraint of the
mobile devices resulting from the requirements for
the mobile devices to be small and light. The battery
power is required not only to run the mobile device
to perform user started operations but also to keep
the device functioning and continuing its quiescent
services. Therefore, an adequate amount of
remaining battery life is essential for the satisfied
device owners.
The battery charges drain at a slow constant rate
(qdr) to keep the device active and functioning.
However, each data communication episode uses
significantly larger amount of battery power to
radiate signals and to perform the related
computational activities. This demand is determined
by a number of factors including the communication
mode, duration and the noise characteristics of the
communication channel. However, drained battery
charge is not a permanent liability. The battery
levels are restored by recharge of the device
batteries. Considering the importance of the
remaining battery charge for the mobile device, a
number of objectives exist in the model.
The remaining electric charge in the battery after
a data communication episode is a function of three
arguments: the state of the battery before the
communication episode, the communication mode
used for the data communication, and the duration of
communication. The communication duration is
primarily determined by the amount of data transfer
and the transmission speed for the communication
mode. However, transmission speed is also affected
by the environmental conditions such as the channel
noise and congestion. Smaller battery discharge in
completing the data communication needs of a
service leaves more charge for other services; thus
,
where
,
∗
(3)
Closely related to this objective are the
constraints that determine the feasibility of the
communication episode. Firstly, the communication
mode should be available over the period delay to
delay + size(data) / speed(mode) for the data to be
communicated. Secondly, the battery should have
enough charge to complete the data communication.
These requirements are expressed in the two
constraints below:
,
,
(4)
and,
∗
(5)
2.4 Minimize Depleted Battery Charge
Encumbrance
If a resource is used for a purpose, it is not available
for the alternate purposes. Economists describe the
idea as an opportunity cost. A similar dilemma is
also faced by the user of a mobile device. Battery
life is an important resource and a mobile device
becomes inoperative once the battery has run down
below a low charge threshold. The savvy mobile
users take significant care to preserve the battery
life. It is not uncommon to notice the mobile users
minimizing the device usage to preserve the battery
charges to the time when the device battery can be
recharged.
SchedulingDataCommunicationbasedServicesonthePersonalMobileDevices
403
This objective is modelled by the area under the
battery charge line plotted against the time. Each
data communication episode reduces the area; and,
this reduction in the area defines the depleted battery
charge encumbrance for the mobile user. The
encumbrance reduces the future opportunities to use
the mobile device due to the reduced remaining
battery charges. As the common mobile devices use
rechargeable batteries, the encumbrance is cleared at
the next battery recharge.
,
where
,
∗
∗
(6)
2.5 Minimizing Device Unavailable
Duration
A mobile device is not fully available to its user
when it is busy in a data communication based
service. This limitation is a consequence of the
constraints on the computational resources as well as
the limited communication bandwidths available to
the mobile devices.
A long access time is annoying and lowers user
satisfaction (Büyüközkan, 2009; Fogelgren-
Pedersen, 2005). A longer access period also drains
more battery charges (Chen, 1999).
The device busy interval is the time interval
required to successfully transmit the data using the
available communication modes – this also is a
period of significantly restricted availability of the
mobile device. Thus, the objective:
,
where
,
(7)
The constraints listed earlier in Equation 5 with
the objective function battery charge depletion apply
to this objective too. However, there is another cause
leading to a potential device unavailable situation.
The events leading to this situation are described as
follows: if the battery charge is depleted by a data
communication episode, the remaining charges on
the battery may not sustain the device in the active
mode till the next recharge time. In this case a period
may occur, before the next recharge, where the
device is unavailable because the device battery
charges have drained to their low threshold level.
The situation may be cared for in the model by
augmenting the constraints on the feasible solutions.
Each feasible decision must ensure:
∗
∗
(8)
If the inequality is not satisfied, the device
battery will discharge to a level below the minimum
threshold and cause the device to become
unavailable. Thus, the function to be minimized
when the condition is not met is:
,
where
,
/
∗//
(9)
3 EXPERIMENTAL
EVALUATION
A number of simulation experiments were
performed to determine the affects of various
operational conditions on the communication
outcomes. Each experimental set-up may be viewed
as consisting of two components -- scenario and
case. The term scenario refers to the general
environmental data such as the list of available
communication modes, the communication mode
availability schedules, and the battery recharge
schedule. These describe operational environment
in which the mobile device is operating irrespective
of specific user activity. The term case is used to
represent individual data transfer requirement. A
case refers to specific activity and is described by
the access time, the estimated size of the data which
is required to be transmitted, the state of battery
charge, and also the festive and the obsolesce
interval of the activity. In order to support the aim of
understanding the relationships between the
decisions and their effects on the outcomes, these
scenarios were organized in order of increasing
complexity. The earlier scenarios were simple and
did not involve any overlapping in the
communication mode availability -- there is only one
ICEIS2013-15thInternationalConferenceonEnterpriseInformationSystems
404
available communication mode at each point in time.
Progressively, the later scenarios included more
options and opportunities to be selected from a range
of alternatives.
Successful execution of simulations based on the
realistic data transfer needs and communication
parameters provide a proof of the concept for our
model. The results of the experiment would offer a
clear and measurable indication of the range of
objective function values which a mobile device user
can observe when using their device. The objective
functions that do not show significant variation in
the Pareto Front across the experiments are unlikely
to affect the user’s decisions to any great degree.
Here we only present one experiment consisting
of a single scenario containing one case. The set-up
data and the outcome results are presented in detail
to give clear view to the reader.
Table 1: Communication mode.
Communication Mode
Cost
($/GB)
Power Use
(mAh/sec)
Wi-Fi Provider X 0.999 0.08
CELLULAR Provider X 19.95 0.03
Wi-Fi Provider Y 0.777 0.08
CELLULAR Provider Y 17.75 0.03
Table 1 presents the communication modes
which are available in this sample scenario. As
presented, there are two service providers in this
scenario, which are Provider X and Provider Y. Both
of them provide Wi-Fi and cellular data network
with a different pricing scheme and availability
schedule. In terms of pricing scheme, Provider X
commands a slightly higher access cost for both Wi-
Fi and cellular data network compared to Provider
Y. Moreover, in this scenario, the power
consumption for each communication mode to
transmit the data is also varying (0.08mAh/sec for
Wi-Fi and 0.03mAh/sec for cellular data network).
The availability schedule for each
communication mode in this scenario is presented in
Table 2. This table represents not only the
availability of each communication mode provided
by the service provider but also the quality of the
communication mode in each particular period of
time (represented by speed fluctuation for each
communication mode). As presented, there are two
periods of time where there is no communication
available to be used. These periods of time are
implemented in this experiment to simulate the dead
zone or the period without any mobile
communication reception.
Table 2: Communication mode availability schedule.
Start End Communication Mode Speed (Mbps)
0:00 2:59 Wi-Fi Provider X 30
3:00 5:59 Wi-Fi Provider X 50
1:30 3:29 Wi-Fi Provider Y 45
3:30 6:29 CELLULAR Provider Y 12
6:00 6:59 CELLULAR Provider X 15
7:00 7:59 CELLULAR Provider X 5
6:30 7:29 Wi-Fi Provider Y 25
7:30 8:29 CELLULAR Provider Y 8
8:00 8:59 Wi-Fi Provider X 30
9:00 9:59 Wi-Fi Provider X 40
8:30 9:29 Wi-Fi Provider Y 25
9:30 10:29 CELLULAR Provider Y 12
10:00 10:59 CELLULAR Provider X 10
11:00 11:59 CELLULAR Provider X 5
10:30 11:29 Wi-Fi Provider Y 35
11:30 11:59 CELLULAR Provider Y 8
12:00 12:59 N/A 0
13:00 13:59 CELLULAR Provider X 15
14:00 14:59 CELLULAR Provider X 10
13:30 14:29 Wi-Fi Provider Y 25
14:30 15:29 CELLULAR Provider Y 4
15:00 15:59 Wi-Fi Provider X 50
16:00 16:59 Wi-Fi Provider X 40
15:30 16:29 Wi-Fi Provider Y 45
16:30 16:59 CELLULAR Provider Y 12
17:00 17:59 N/A 0
18:00 18:59 Wi-Fi Provider X 30
19:00 19:59 Wi-Fi Provider X 40
18:30 19:29 Wi-Fi Provider Y 45
19:30 20:29 CELLULAR Provider Y 8
20:00 20:59 CELLULAR Provider X 10
21:00 21:59 CELLULAR Provider X 15
20:30 21:29 Wi-Fi Provider Y 25
21:30 22:29 CELLULAR Provider Y 12
22:00 22:59 Wi-Fi Provider X 30
23:00 23:59 Wi-Fi Provider X 50
22:30 23:29 Wi-Fi Provider Y 35
23:30 23:59 CELLULAR Provider Y 8
There are three battery recharge periods in this
scenario as presented in Table 3. These periods of
time are introduced in the experiment in order to
simulate the period for a user to recharge his/her
mobile device.
SchedulingDataCommunicationbasedServicesonthePersonalMobileDevices
405
Table 3: Battery recharge schedule.
Start End
13:30 13:44
17:30 17:44
21:30 21:44
A sample case is presented in this paper in order
to run the simulation based on the sample scenario
described earlier. As mentioned before, the
parameters in a case is used to represent the situation
of data transfer. In this case, the parameters are
presented in Figure 1.
Figure 1: Data transmission parameters.
As presented, the intended user’s data transfer in
this case is at 11:50 and the estimated size of the
data which is going to be transmitted is 700MB. The
data has a festive interval of 180 minutes (3 hours)
and an obsolesce interval of 240 minutes (4 hours).
In this case, the mobile device which is going to be
used to transmit the data has 200mAh of battery
charge. This information is going to be used as the
input parameters in order to find the set of solutions
that can produce the optimum level of satisfaction.
In order to solve the problem in multi-objective
optimization, there are several evolutionary
algorithms available and can be implemented. For
our model, Non-dominated Sorted Genetic
Algorithm version 2 (NSGA II) is the one which
chosen to run the simulation. Compared to the first
version of NSGA, the NSGA II is more efficient in
terms of computational process. NSGA II uses
elitism and a crowded comparison operator to
maintain diversity of the population. Elitism is
implemented in NSGA II in order to help achieve
better convergence. In terms of converging near the
Pareto Front and in terms of maintaining diversity
among obtained solutions, NSGA II in general is
better than PAES (Pareto Archived Evolution
Strategy) and SPEA (Strength Pareto Evolutionary
Algorithm), the two other elitist multi-objective
evolutionary algorithms (Deb, 2002; Deb, 2010;
Coello, 2007).
As the building blocks to construct the
simulation model, jMetal framework is used. jMetal
stands for Meta-heuristic Algorithms in Java, it is a
framework for constructing and solving the multi-
objective optimization problem using evolutionary
algorithms. The framework is based on Java
programming language and has been used in a wide
range of applications since it was built as an easy to
use, flexible, and extendable software package. The
ease of use, flexibility, and extendibility can be
achieved by jMetal since it takes full advantage of
the capabilities that Java offers and is structured in a
way that a problem can be developed as an
independent class from the algorithm that solves it.
A wide range of core classes which can be used as
the building blocks of multi-objective meta-
heuristics are provided by this framework in order to
take advantage of code-reusing. In addition, the
evolutionary multi-objective algorithms in this
framework are tested for their performance with
standard multi-objective optimization problems
(Durillo and Nebro, 2011.).
Based on these input parameters as presented in
Figure 1, a set of non-dominated/optimum solutions
is produced and presented in Figure 2. A number of
non-dominated solutions that were produced and
reported by the simulator and in Figure 2 do not
seem to be different from each other especially in
Delay period from 100 to 110 minutes. This may be
due to close real values for the metrics that exceed
the others by only a very small amount. However,
we like to stress that these nearly similar outcomes
provide assurance that the outcomes are not very
sensitive to the delay and mode value in this period -
- the user can select delay in a relaxed calm fashion.
Figure 2: A set of optimum solutions.
As shown in Figure 2, the optimum solution with
the highest efficacy level can be achieved by
postponing/delaying the data transmission for
104.2277, 104.8076, 105.3889, 107.8175, 108.1544,
or 109.7744 minutes (time period 13:34:13 to
15:48:04) and using Wi-Fi provided by Provider Y
as the communication mode to transmit the data. As
ICEIS2013-15thInternationalConferenceonEnterpriseInformationSystems
406
Figure 3: Blind solutions and optimum solutions
comparison.
presented in Table 2, in this period of time there are
more than one communication modes available, and
in order to produce optimum satisfaction level, Wi-
Fi provided by Provide Y is chosen.
By postponing the data transfer for 190.3732
minutes and using Wi-Fi provided by Provide X,
another optimum satisfaction can be achieved. This
optimum solution offers the shortest access interval
compared to the others optimum solutions in this
case.
As an evaluation, we also compare these
optimum solutions with the possible blind solutions
which are a user might chose for this case. A blind
solution is a solution where the device owner
performs the data communication activity
immediately. Most mobile device data
communication today operates in this mode. Based
on Table 2, there are two available communication
modes for blind solution in this case: cellular data
network provided by Provider X or cellular data
network provided by Provider Y. We compare these
blind and optimum solutions based on the five
objective functions which we have explained earlier.
The following figures show the solutions
comparison for this case.
As shown in the Figure 3, the optimum solutions
can significantly produce better outcomes in order to
satisfy the five objective functions which we
discussed earlier. Blind solution can only produce a
better outcome in terms of efficacy due to immediate
data transmission. A mobile device owner can
achieve best outcomes for data communication
activity by choosing a delay and comparing the
various outcomes against those that are achieved at
other delays.
4 CONCLUSION AND FUTURE
WORK
The paper has presented a multi-objective model for
scheduling the data communication based services
on the mobile devices. As these objectives are
incomparable with each other and represent different
needs they cannot be assigned meaningful weights to
generate a single outcome. Even if such weights
were feasible, each mobiles user will attach different
weight of importance to these objectives. Indeed,
one would expect the same user to assign different
weights to these objectives at the different times.
Pareto optimal solutions provide the best option for
determining efficient schedules.
Using realistic parameter values and cyclic
lifecycle of the mobile users we have run several test
SchedulingDataCommunicationbasedServicesonthePersonalMobileDevices
407
scenarios. The experiments support the common
wisdom of preferring a wired data communication
mode over a Wi-Fi mode and preferring the Wi-Fi
communication over the cellular communication.
This rule receives support as these preferred modes
also frequently overlap with the battery recharge
opportunities.
The increasing adaption of the cloud
computation and resources would make the services
on the mobile devices even more dependent on the
externally located data. The models for scheduling
the data communication would optimize the use of
mobile device resources and capabilities.
There are several possibilities that can and need
to be explored as future works. Our model can be
augmented in order to handle multiple transfers and
multiple data access. The enhanced model may be
used to schedule the data transfer in an optimized
way. Alternatively one may propose an algorithm to
prioritize the transfers and choose a subset of them
for actual transfer consistent with the expected cost
and benefit outcomes.
A machine learning capability can be integrated
into this model to predict the mobile user’s daily
activity in order to generate a more accurate
communication availability schedule. A simulator
tool may also be of value to monitor the device and
its usage to provide accurate and precise estimates of
various parameters used in the model.
ACKNOWLEDGEMENTS
The work was part of the first author’s Master thesis
done at School of Computing and Information
Systems, University of Tasmania, Australia (2012).
REFERENCES
Ballard, B., 2007. Designing the mobile user experience,
Wiley.
Blechar, J. et al., 2006. Exploring the influence of
reference situations and reference pricing on mobile
service user behaviour. In European Journal of
Information Systems, 15 p.285–291.
Büyüközkan, G., 2009. Determining the mobile commerce
user requirements using an analytic approach. In
Computer Standards & Interfaces, 31 (1), p.144–152.
Chen, J. et al., 1999. Performance comparison of battery
power consumption in wireless multiple access
protocols. In Wireless Networks, 5 (6), p.445-460.
Cui, Y. and Roto, V., 2008. How people use the web on
mobile devices. In Proceedings of the 17th
international conference on World Wide Web. 1st ed.
Beijing: ACM, p.905-914.
Coello, C. A. et al., 2007. Evolutionary algorithms for
solving multi-objective problems, Springer. New York.
Constantiou, I. D. et al., 2007. The four incremental steps
towards advanced mobile service adoption. In
Communication of ACM, 50 (6), p.51-55.
Deb, K. et al., 2002. A fast and elitist multi-objective
genetic algorithm: NSGA-II. In Evolutionary
Computation. IEEE Transactions, 6 (2), pp.182-197.
Deb, K., 2010. Multi-objective optimization using
evolutionary algorithms, Wiley.
Durillo, J. and Nebro, A., 2011. jMetal: A Java framework
for multi-objective optimization. In Advances in
Engineering Software, 42 (10), p.760-771.
Fogelgren-Pedersen, A., 2005. The Mobile Internet: The
Pioneering Users’ Adoption Decisions. In Proceedings
of the 38th Annual Hawaii International Conference
on System Science, p.84b.
Laplante, P., 2004. Real-time systems design and analysis:
an engineer's handbook, IEEE.
Sarker, S. and Wells, J., 2003. Understanding Mobile
Handheld Device Use and Adoption. In
Communications of ACM, 46 (12), p.35 - 40.
Williams, R., 2006. Real-Time systems development,
Oxford.
ICEIS2013-15thInternationalConferenceonEnterpriseInformationSystems
408
