ALGORITHM FOR ENERGY-EFFICIENT AND AUTOMATIC

CONFIGURATION OF PEDESTRIAN NAVIGATION

SERVICES: SMARTPHONES ON CLOUDS

Monir H. Sharker and Hassan A. Karimi

University of Pittsburgh, Geoinformatics Laboratory, School of Information Sciences, Pittsburgh, United States

Keywords: Pedestrian Navigation Services, Smartphone, Cloud Computing, Energy-Efficient Configuration, Load

Balancing.

Abstract: Smartphones have been playing a major role as personal navigation aids to pedestrians, among other

travelers. Pedestrian Navigation Services (PNSs) on smartphones face challenges such as limited battery

power, relatively lower computational speed, limited storage capacities, and varying quality of available

networks. Automatic configuration of PNS components is very important to address these challenges. In this

paper, we present a novel approach for optimum configuration of components and computations of PNSs

between smartphones and cloud platforms. An algorithm is developed based on three models, Minimum

Computation (MinComp), Minimum Communication (MinComm), and Balanced Computation-

Communication (BalCC), to allow smartphones to operate navigation services optimally in a distributed

environment over cloud platforms. Each model was simulated by using various values for each parameter.

1 INTRODUCTION

Advancements of key technologies including geo-

positioning sensors, wireless communications,

mobile devices, and the internet have been paving

the way for development of Pedestrian Navigation

Services (PNSs) on smartphones (Karimi, 2011). To

better understand what PNSs are and how they work,

we make a distinction between navigation systems

and navigation services where the former is a

reference to stand-alone systems, with all data stored

locally and all computations performed in a

centralized manner, and the latter is a reference to

services with data and computation distributed over

a network of computers. We define PNSs as those

navigation services available on smartphones that

can assist pedestrians to move from one place to

another and can meet their specific needs and

preferences.

While from a user’s perspective, little

differences can be observed between navigation

systems and services, from a developer’s

perspective, there are challenges in developing

navigation services, in particular PNSs. In PNSs,

challenges are stemmed from dynamic

characteristics of mobile devices such as limited

battery power, relatively lower computational speed,

limited storage capacity, and varying quality of

available networks signal strength. PNSs are based

on client-server architectures, where clients are

increasingly smartphones and servers are service

providers. Optimum distribution of data and

computation over client-server environments, e.g.,

energy efficient services, requires a new algorithm

that can automatically configure service components

between the clients and the server. In this paper, we

present an algorithm that allows PNSs on

smartphones to be configured dynamically and

automatically based on specific criteria where

servers are cloud computing platforms. Service

providers are increasingly resorting to cloud

computing platforms primarily to address scalability

of their services and to minimize services costs.

Unique aspects of PNS components and features

require special attention to configure those in a

manner aiming for energy-efficient and compute-

efficient services. For example, some existing

navigation systems and services provide

navigational assistance to pedestrians by using road

networks, as they are used for assisting drivers.

Road networks are unsuitable and ineffective for

assisting pedestrians with their mobility as

H. Sharker M. and A. Karimi H..

ALGORITHM FOR ENERGY-EFFICIENT AND AUTOMATIC CONFIGURATION OF PEDESTRIAN NAVIGATION SERVICES: SMARTPHONES ON

CLOUDS.

DOI: 10.5220/0003953400850090

In Proceedings of the 1st International Conference on Smart Grids and Green IT Systems (SMARTGREENS-2012), pages 85-90

ISBN: 978-989-8565-09-9

 2012 SCITEPRESS (Science and Technology Publications, Lda.)

pedestrians usually walk on sidewalks and not on

road center lines. For effective navigation assistance

to pedestrians, PNSs must contain pedestrian paths

in the underlying navigation environments. The core

data in a pedestrian path database is a network of

sidewalks which contains geometry, topology, and

attributes of the pedestrian path (Kasemsuppakorn

and Karimi, 2009). Geometrical data in a pedestrian

path database include coordinates of decision

points—a decision point is a juncture whereby a

decision must be made by the pedestrian (e.g., an

intersection)—and of intermediate points on

sidewalks. Topological data in a pedestrian path

database include data about connectivity between

decision points and sidewalks. Attribute data in a

pedestrian path database includes data such as

sidewalk name, sidewalk width, and sidewalk type.

Given all these PNS components and features,

the limitations of mobile devices, diversity of user

requirements, and the environmental factors, a

dynamic configuration scheme is a key requirement

for providing efficient services. For example, given

a complex computational task and a mobile device

with low battery power, it is best to have the task

performed on the server side. On the other hand, a

simple computational task that is required frequently

is best to be performed in the client side to minimize

communication overheads. Many combinations,

considering device constraints and computations at

hand, among other factors, are possible for which an

efficient solution is desirable.

The contribution of the paper is an algorithm for

automatic configuration of data and computation

components in PNSs. The algorithm addresses many

situations where minimizing energy consumption is

the highest priority. The rest of the paper is

organized as follows. In Section 2, the main

components of PNSs and possible model

configurations are discussed. Section 3 describes the

algorithm. The models and the simulation of the

algorithm are explained in Section 4. Simulation

results are discussed in Section 5. Section 6

highlights related works and Section 7 concludes the

paper and discusses future research.

2 COMPONENTS AND MODELS

The main components and functions of PNSs

include sidewalk network map database, geocoding,

map matching, routing, smartphone (GPS-enabled),

and cloud computing server and/or service provider.

Sidewalk network map databases are the main data

source for PNSs. Geocoding is the process of

assigning geographic coordinates (lat/long) to a

given place by comparing its description to the

descriptions of location-specific elements in the

reference datasets (Goldberg, 2007). Map matching

is a technique for matching pedestrian trajectories to

correct pedestrian paths in the sidewalk network

(Karimi et al., 2006). Routing computes a preferred

path from any given origin to destination

(Kasemsuppakorn and Karimi, 2009). Proposed

models of PNSs are explained bellow in brief.

2.1 Minimum Computation (MinComp)

For a resource poor client device, the MinComp

model is considered where most computations are

submitted to the cloud. In the MiniComp model,

with the assumption that the device battery is low,

its computation capability is limited, and its

available storage is low, all computational tasks are

submitted to the cloud. In this model, the client is

mainly responsible for submitting the user request to

the cloud, receiving the response from the cloud

upon task completion, and presenting the results to

the user.

2.2 Minimum Communication

(MinComm)

In the MinComm model, most computations are

performed at the client device provided that the

client has sufficient battery power, computational

capability, and storage capacity. This model is

suitable in situations where the device can handle

the computational load and the user needs fast

response. This model is also suitable in situations

where network availability is limited and network

quality is poor. The main responsibilities of the

cloud in this model are map storage, map rendering,

and map data transfer.

2.3 Balanced Computation-

communication (BalCC)

The third model, BalCC, is the most suitable model

for many clients since client devices fall into this

category most often. This is basically a trade-off

between the MinComp and MinComm models. In

situations where the client device has an average

battery power, average computation capability,

moderate amount of storage capacity, and network

availability is in moderate level, the BalCC model

provides the most optimal solution. In this model,

some of the computations are performed in the client

side while some in the cloud maintaining a balance

SMARTGREENS2012-1stInternationalConferenceonSmartGridsandGreenITSystems

between the client and the server. The flexibility

with the BalCC model is that it chooses the optimum

load balance both for the client device and the cloud

based on the dynamic status of the resources on the

fly.

3 ALGORITHM

The algorithm has a decision engine where the

decision, about which model to choose on the fly, is

made. To make a decision, the algorithm (see Figure

1) consults parameters status such as battery power

(BP) level, storage capacity (SC) level, processor

speed (PS) of the client device, and the

environmental factors like network availability

(NA).

Battery power (BP) is the most important

parameter considered by the algorithm. The

objective is to minimize energy consumption.

However, once the algorithm recognizes that the

client device has high level of energy, it takes the

other parameters into account. In situations where

power is in medium or low range, the algorithm

gives BP the highest priority to minimize energy

consumption.

Each device’s parameter responds by providing

its level of strength by means of a score in

accordance with the probability of the status. By

accumulating all the response scores, the algorithm

generates a scaled score and sends the score to the

decision engine. The decision engine decides the

model, with the highest score, to be executed.

The algorithm has the following steps.

1. Obtain specifications of the models.

2. Acquire the client device status

probabilities along with a scaled score for

all parameters (BP, SC, PS, and NA).

3. Monitor continuously the current status of

the device and feed the latest status score.

4. Calculate the overall score (expected value)

for each model taking all the possibilities

Table 1: PNS model specifications, decision criteria along with their pros and cons.

Model

Functions on

Smart phones

Functions on

Cloud Servers

Adaptation

Criteria

Pros Cons

Minimum-

Computation

• GPS data acquisition

• Get user input

source/destination

/options

• GPS data/user input

transfer (send)

• Map data transfer

(receive, cache)

• Map match data

transfer (receive)

• Map data update

•

GPS data

transfer (receive)

• GPS data map

match

• Map extent

retrieval

• Geocode user

data

• Route computing

• Map rendering

• Map data

transfer (send)

• Power/energy

consumption

• Cost: pay per

usage

• Computation

capability of

client

• Minimum

storage

• No burden of

updates on

Smart phones

• Privacy

compromised

• Subject to

Internet speed

and interruption

• Frequent GPS

data update

Balanced

Communication-

Computation

• GPS data acquisition

• User input

source/destination

/options

• Map data transfer

(receive, cache)

• Geocode user data

• GPS data map match

• Map match data

transfer (send)

• Map data update

• Map match data

transfer (receive)

• Route computing

• Map extent

retrieval

• Map rendering

• Map data

transfer (send)

• GPS Data update

• Faster

• Increased

utilization

• Loads are

balanced

• Faster

response

• Synchronization

could be an issue

• Frequent/Modera

te GPS data

update

Minimum-

Communication

• GPS data acquisition

• GPS data

transfer(send)

• Map data transfer

(receive)

• GPS data map match

• User input

source/destination

/options

• Geocode user data

• Map extent retrieval

• Route computing

• Map rendering

• GPS data

transfer (receive)

• Map retrieval

• Map data

transfer (send)

• Minimum

Internet

interruption and

disconnection

• Cost: One-time

fee for

installation

• No/Limited

privacy

compromised

• Infrequent map

data download

• High computing

demand on

Smartphones

• Large storage on

smartphones

ALGORITHMFORENERGY-EFFICIENTANDAUTOMATICCONFIGURATIONOFPEDESTRIANNAVIGATION

SERVICES:SMARTPHONESONCLOUDS

into consideration.

5. Select the model with the highest score.

6. Continue the process upon request.

Figure 1: Flowchart of the algorithm.

4 MODELS AND SIMULATION

4.1 Models

To calculate the overall score (expected value) for

each model, for one set of probabilities, the

following equations are used:

)1()()()()()()(

AScaleValNAPPSPSCPBPPMinCompScore

ijkl

××××=

∑∑∑∑

====

)2()()()()()()(

BScaleValNAPPSPSCPBPPMinCommScore

ijkl

××××=

∑∑∑∑

====

)3()()()()()()(

CScaleValNAPPSPSCPBPPBalCCScore

ijkl

××××=

∑∑∑∑

====

where, p

(bp) is the probability of battery power

of level i, i= 1,3 (high, medium, low), p

(sc) is the

probability of storage capacity of level j, j= 1,3

(high, medium, low), p

(ps) is the probability of

processing speed of level k, k= 1,3 (high, medium,

low), and p

(na) is the probability of network

availability of level i, i= 1,3 (excellent, good, poor).

The scale value (ScaleVal) is determined through a

Bayesian belief network based on the parameters in

each model.

The above score calculation, for each model, is

only for one set of probabilities of the parameters.

All possible combinations of probability set are then

generated and the score for each model is calculated

for all possibilities. Thus, comparing the overall

score generated for the models for a given set of

probability, the model with the highest score is

chosen.

4.2 Simulation

We used influence diagram, as shown in Figure 2, to

simulate the algorithm. The decision variable is set

as Model Decision having three values MinComp,

MinComm, and BalCC. Three variables,

Battery_Power, Storage_Capacity, and

Processor_Speed, at three levels of probabilities,

High (0.6), Medium (0.3), and Low(0.1) for each, are

considered. A fourth variable, Network_Availability,

also has three levels of probabilities as Excellent

(0.6), Good (0.3), and Poor (0.1). The levels of

variables could be chosen to any level with

appropriate scaling factor. We simulated all possible

outcomes for each model for all possible input sets.

Then comparing the model outcomes for each input

set, the decision model with the highest score

suitable for the given situation was selected. We

used GeNIe (GeNIe 2011), to create and evaluate a

decision theoretic model for PNSs.

5 RESULT AND DISCUSSION

We used three levels of probabilities for each

variable, which means 3!=6 sets of probabilities for

each variable. In total, there are 6x6x6x6=1296

possible different decision points, considering all

four variables together. The score for each model

was calculated and a decision was made. The

possibilities of Storage_Capacity (SC) are shown in

Figure 3. We used a scale value ranging from 2.0 to

10.0 for each model. The final score calculated for

each model hence ranges from 2.0 to 10.0.

For one example set of combination, i.e.,

BP[High(0.1), Medium(0.3), Low(0.6)],

SC[High(0.1), Medium(0.3), Low(0.6)],

Start

Read Client Status

Battery_Power,

Storage_Capacity,

Processor_Speed,

Network_Availability

Update Status

Read Models

A,B,C

Calculate Model Score

Score(A),

Score(B),Score(C)

MS=Score(A)

A-> Minimum Computation

B-> Minimum Communication

C-> Balanced Compute-Communicate

Run model B

MS=Score(B)

Run model A

Run model C

Continue?

Stop

MS=max[Score(A),

score(B), Score(C)]



SMARTGREENS2012-1stInternationalConferenceonSmartGridsandGreenITSystems

Figure 2: Influence diagram showing simulation result for

an example set of combination.

Figure 3: Possible states of the chance variable,

Storage_Capacity.

PS[High(0.6), Medium(0.3), Low(0.1)], and

NA[Excellent(0.6), Medium(0.3), Low(0.1)], the

calculated scores are: MinComp = 7.6, MinComm =

4.4, and BalCC = 5.61. The result for this specific

case is verified by Genie (Bayesian Network) as

shown in Figure 2. The complete simulation result is

plotted in Figure 4 for all the 1296 decision

alternatives. This result gives an idea about how the

decision switches between the models over the

change of probabilities of variables. We will conduct

the experiment on a real PNS in next phase.

6 RELATED WORKS

Fundamental challenges for mobile computing, due

to limited resources on mobile devices, poor

wireless network, and limited energy sources are

outlined by Satyanarayanan (1996). The state-of-the-

art cloud computing paradigm is analysed by

Buyya

et al. (2009, 2011) where cloud computing is

recognized as the technology to offer computing

services anywhere in the world on demand. In a

distributed mobile computing framework, to

minimize the power consumption for the client,

computation loads are shared among the nodes such

that the required energy can be fairly used as

contributed by Liang et al. (2007). Without a priori

knowledge of the application, the optimal

deployment, that minimizes CPU usage at the

mobile device keeping the used bandwidth minimal,

is calculated (Verbelen et al., 2010). Banavar et al.

(2000) proposed a new application model for

pervasive computing that gives techniques such as

device-specific rendering, application apportioning,

and application adaptation. Hao-Hua Cho et al.

(2004) describe the challenges of constructing a

new, wireless, web-service architecture, based on a

client model.

Stoer and Wagner (1997) describe an algorithm

that finds the minimum cut to divide a graph in two

partitions. The approach here is that to calculate the

optimal partitioning of the components, it represents

the services as a weighted graph, and transforms the

deployment problem into a graph partitioning

problem. Han et al (2008) present graph partitioning

algorithms aimed specifically at the problem of

partitioning software for mobile devices. One of the

key considerations is an adaptor which monitors

system-level states, such as CPU load or network

bandwidth. They use the adaptors to adjust

application parameters or activate functional

reconfiguration actions based on a fuzzy control

model.

Figure 4: Simulation result for all alternatives.

However, none of these research projects takes

all the parameters mentioned in this paper into

account and considers all possible combinations.

Our proposed algorithm contributes by using a

Bayesian belief network (influence diagram) to find

an optimal configuration. For any combination of

ALGORITHMFORENERGY-EFFICIENTANDAUTOMATICCONFIGURATIONOFPEDESTRIANNAVIGATION

SERVICES:SMARTPHONESONCLOUDS

parameters, the influence diagram generates an

overall score (see Figure 2 for an example

combination) based on which the decision is made.

Accordingly, each function in PNSs is executed

either on the client or on the server depending on

which provides the most optimal performance for

the situation at hand. This algorithm also utilizes

resources on user’s device optimally and more

effectively.

7 CONCLUSION AND FUTURE

WORK

This paper describes different configurations

explaining all possible computations and

communications models in PNSs using smartphones

(clients) and clouds (servers). An algorithm that

analyses these models, in order to automatically

decide a suitable configuration, is presented. The

algorithm was simulated and the result shows that

the algorithm provides the most optimum model

given the environmental and device parameters of

PNSs. The algorithm is dynamic and reconfigurable.

Once the functions, specifications of power usage,

storage capacity, processing capabilities of the

mobile device, and the network provider information

are available, this algorithm will automatically

decide on the most suitable model for the scenario.

Future work will include implementation and

deployment of the algorithm on mobile devices and

cloud computing platforms.

REFERENCES

Banavar, G., Beck, J., Gluzberg, E., Munson, J., Sussman,

J., and Zukowski, D. (2000). Challenges: An

Application Model for Pervasive Computing. 6th

ACM/IEEE International Conference on Mobile Net-

working and Computing, Boston, USA: ACM Press,

pp. 266-274.

Buyya, R, Yeo, C. S., Venugopal, S. Broberg, J., and

Brandic, I. (2009). Cloud computing and emerging IT

platforms: Vision, hype, and reality for delivering

computing as the 5th utility. Future Generation

Computer Systems, 25(6):599.

Buyya, R., Broberg, J., and Goscinski, A. (2011). Cloud

Computing: Principles and Paradigms. John Wiley &

Sons, Inc., NJ.

GeNIe (Graphical Network Interface) (2011).

http://genie.sis.pitt.edu/about.html

Goldberg, D.W., Wilson, J.P. and Knoblock, C.A., (2007).

Fromtext to geographic coordinates: the current state

of geocoding. Journal of the Urban and Regional

Information Systems Association, 19, 33–46.

Han, S, Zhang, S, Cao, J., Wen,Y., and Zhang Y. (2008).

A resource aware software partitioning algorithm

based on mobility constraints in pervasive grid

environments. Future Generation Computer Systems,

24(6):512-529.

Hao-hua, C., Chuang-wen, Y., and Chao-ming, T., (2004).

Challenges: wireless Web services. 10

International

Conference on Parallel and Distributed Systems

(ICPADS 2004), IEEE Computer Society, pp. 657.

Karimi, H. A., Conahan, T., and Roongpiboonsopit, D.

(2006). A Methodology for Predicting Performances

of Map-Matching Algorithms. 6th International

Symposium on Web and Wireless Geographical

Information Systems (W2GIS 2006). Hong Kong,

December 4-5.

Karimi, H. A. (2011). Universal Navigation on

Smartphones. Springer.

Kasemsuppakorn, P. and Karimi, H. A. (2009).

Personalized Routing for Wheelchair Navigation.

Journal of Location Based Services, Vol. 3, No. 1, pp.

24-54.

Liang, W.Y., Hsieh, Y.M., and Lyu, Z.Y. (2007). Design

of a dynamic distributed mobile computing

environment. International Conference on Parallel

and Distributed Systems, IEEE Explore.

Satyanarayanan, M. (1996). Fundamental Challenges in

Mobile Computing. 15

ACM Symposium on

Principles of Distributed Computing, pp. 1-7.

Stoer M. and Wagner F., (1997). A simple min-cut

algorithm, J. ACM, 44(4):585-591.

Verbelen, T., Hens, R., Stevens, T., and De Turck, F.

(2010). Adaptive online deployment for resource

constrained mobile smart clients. 3

International

ICST conference on MOBILe Wireless MiddleWARE,

Operating Systems, and Applications(Mobilware).

SMARTGREENS2012-1stInternationalConferenceonSmartGridsandGreenITSystems