3056

IEEE TRANSACTIONS ON PARALLEL AND DISTRIBUTED SYSTEMS,

VOL. 25,

NO. 12,

DECEMBER 2014

A Context-Aware M2M-Based Middleware

for Service Selection in Mobile Ad-Hoc Networks
Nusrat Ahmed Surobhi, Student Member, IEEE and Abbas Jamalipour, Fellow, IEEE
AbstractThis paper proposes a novel middleware for service selection in mobile ad-hoc networks with a particular focus on
scenarios immediately and subsequently afterwards an emergency. The proposed middleware, operating on a mobile users hand-held
device and collecting the users contexts through machine-to-machine connectivity, has three major contributions as compared to the
current literature. The middleware, based on the collected contexts, firstly classifies a service request, e.g., safety-related or comfort
service. Then, it initiates the required network connectivity for service discovery, i.e., the ad-hoc connectivity when an infrastructurebased network is inaccessible. Finally, the middleware selects the service based on a method specifically proposed for that service
category, including pre-defined realistic user contexts, to access it. The simulation results show that the middleware achieves up to
12 percent higher success rate compared to the minimum hop count service selection method. Compared with the same method, it has
a response time up to 50 percent lower for the safety-related but 80 percent higher for the comfort services. Yet, the middleware attains
up to 96 percent user satisfaction rate for both service classes.
Index TermsWireless communication, mobile environments, service-oriented architecture

INTRODUCTION

N the last decade, wireless networking has become one

of the most dominant ways of communication using
the advances, availability and built-in inexpensive wireless connectivity such as Wi-Fi [1]. In many public and
office environments, these wireless networks are usually
part of infrastructure-based networks, for example, wireless local area networks (WLANs). In WLANs, a centralized server is responsible to provide the mobile users
with safety-related services, e.g., location of assembly
points in a campus, emergency phone numbers, etc.
Besides, it is also responsible to provide the Internet to
those users such that they can access their everyday comfort services like comparing fuel prices, peer to peer content exchange, finding daily stock quotes, etc. [2].
However, communications with the central server
can become disrupted when the network infrastructure
becomes entirely/partially damaged by an emergency, e.g.,
a natural disaster like hurricane Sandy [3]. As a result,
many users may not be able to directly access services from
the server [4]. While some of these users may continue to
access services utilizing the ad-hoc connectivity established
over their mobile devices, they are likely to become
completely disconnected from the server with the increased
user mobility during evacuation. Subsequently, the user initiated mobile ad-hoc networks (MANETs) become one of
the most preferred ways to discover and access the safetyrelated services [5]. Additionally, they are also proven to be

The authors are with the School of Electrical and Information Engineering,
University of Sydney, NSW 2006, Australia.
E-mail: {ns.ahmed, abbas.jamalipour}@sydney.edu.au.

Manuscript received 11 Sept. 2013; revised 28 Nov. 2013; accepted 30 Dec.

2013. Date of publication 23 Feb. 2014; date of current version 14 Nov. 2014.
Recommended for acceptance by N. Kato.
For information on obtaining reprints of this article, please send e-mail to:
reprints@ieee.org, and reference the Digital Object Identifier below.
Digital Object Identifier no. 10.1109/TPDS.2014.2307875

helpful to discover and access the everyday comfort services in the case of a longer communication disruption
caused by an extreme emergency [6].
Often, users in MANETs are required to implement a
selection strategy since a single service can be offered by
multiple providers especially in service rich urban terrains.
Yet, the commonly used service selection methods for
MANETs [7], [8], [9] fail to distinguish between safetyrelated and comfort services. Moreover, it can be understood that users service expectations are different for each
of these classes. This is due to the fact that contexts, while a
service is requested, play a key role in determining users
expectations [10], [11]. For instance, in reality, users have
been observed to expect safety-related services to be highly
delay-sensitive in nature and thus expect to access them
from one of the most reliable service providers. On the other
hand, the same users are usually observed not to have the
delay-sensitive requirements for comfort services provided
that they can access these services from one of the most popular service providers comparing preceding service consumers opinions published through social networking,
micro-blogging sites etc. [12], [13], [14].
On the other hand, the current mobile devices such as
mobile phones, tablet and laptops are sensor-enabled,
equipped with built-in global positioning system (GPS) and
closely carried by their human users. As a result, these devices are capable of capturing users context information
including users location, velocity, body temperature and
surrounding ambiance. Additionally, these devices can also
monitor and collect other users context information using
their machine-to-machine (M2M) connectivity, e.g., Wi-Fi
connectivity, with those users mobile devices. These context information can be used to classify the requested service and to tailor a service selection strategy specifically for
that class reflecting the users service expectations. Consequently, this work proposes a context-aware M2M-based

1045-9219 2014 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission.
See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.

SUROBHI AND JAMALIPOUR: A CONTEXT-AWARE M2M-BASED MIDDLEWARE FOR SERVICE SELECTION IN MOBILE AD-HOC NETWORKS

middleware which operates on a mobile users device for

service selection in MANETs. The novel aspects of the proposed middleware are outlined below:

The proposed middleware runs on a user device, i.e.,

an application running on a smart mobile phone,
tablet, laptops, etc., in a real network and thus can
capture the users contexts. It collects physical contexts such as user mobility information to classify a
service request through the M2M communications
between the mobile devices carried by the users. The
use of the M2M connectivity makes operations of
information collection and service classification efficient because of requiring lesser resources [15].

The middleware also initiates a MANET automatically over the connectivity enabled user devices for
service discovery in cases when the responsible
server (realized as part of a WLAN-based M2M network) is inaccessible during and after an emergency.

Finally, for service selection, the middleware incorporates a users realistic expectations, which have
been pre-defined within the middleware, for a particular class of service. Hence, for a safety-related
service, the middleware selects one the most reliable
providers by optimizing the hop distance and the
route expiry time (RET) for accessing the service. On
the other hand, for a comfort service, the middleware
selects one of the most popular providers including
preceding users opinions about those service providers and the hop counts to access it.
Note that to the authors best knowledge, such contextaware M2M-based middleware for service selection in
MANETs has not been investigated yet. However, an opinion-based service selection method for comfort services in
MANETs is presented in the authors earlier work in
[16]. That method, however, has not been part of any context-aware selection middleware. Therefore, it requires
manual detection of the comfort services and implementation. Hence, it is inconvenient and inefficient in mobile
environments.
The rest of the paper is organized as follows. Section 2
presents the related works. The preliminaries of the system
model are discussed in Section 3. Section 4 describes the
proposed service selection middleware in detail. The simulation results are presented and discussed in Section 5.
Lastly, Section 6 concludes the paper.

RELATED WORK

While an extensive research has been conducted on the

service discovery in MANETs, only a few works have
investigated the service selection. Broadly, service selection methods are classified into user assisted and automatic service selection [9]. The user assisted selection
methods require users manual efforts for scrutinizing all
discovered providers in order to select the service that
best satisfies the users needs. Since the mobile devices
small screen size, limited battery life and dynamic nature
pose several limitations on such service selection, the
automatic service selection methods are more widely
used. They only require initial user supervision in terms
of setting user profiles and preferred selection

3057

parameters. The discovered providers are then ranked

based on those parameters in order to select the best
ranked provider.
Among the automatic service selection methods, the
most common and widely used service selection method
in MANETs is the minimum hop count (MHC) method.
This method uses a route specific metric, distance
between a user and all discovered service providers, to
select the nearest provider to the user [7], [8]. Additionally, authors of [9] have proposed a service selection
method that is based on a route specific metric as well as
a service specific metric. While the route specific metric
in this work is also based on hop counts, the service specific metric is based on the remaining energy of the service providers. Besides, work in [17] selects a provider
based on a combination of route and service specific metrics, e.g., hop counts and the nominal capacity of the provider respectively. Recently, cross-layer service discovery
and selection in MANETs have attracted much attention
from the research community. The cross-layer service discovery is first proposed in [18] which integrates service
discovery in the routing layer. Following this, works in
[8] and [9] have proposed cross-layer service selection
methods. The cross-layer service selection method proposed in [8] is based on the dynamic source routing and
destination-sequenced distance vector routing protocols.
By implementing the principle of MHC selection method
to this cross-layer service selection method, the authors
has claimed that the proposed cross-layer service selection outperforms the application-layer service selection
methods. On the other hand, work in [9] uses the Ad Hoc
On Demand (AODV) routing protocol to propose a crosslayer service selection method. Regardless of the routing
technique used, such cross-layer service discovery and
selection methods help users to have simultaneous
knowledge of available service providers and route information. As a result, these methods reduce costs and raise
the success of the selection.
Until very recently, service selection is performed assuming a pre-defined state of the underlying network [19].
Therefore, these methods only work well in the targeted settings. Since the nature of MANETs is dynamic, selection
methods that include context information can improve network performances [20]. Work such as [21] is one of the earliest work employing the context information, e.g., location
of the service provider, for service selection. Later, works
such as in [22] has used advanced context information to
select services by including dynamic as well as static attributes of service providers. This work uses a weighted vector
based aggregation function to combine the attributes and
returns the top ranked results to the user. However, this
work is unable to implement any semantic search, match
and reasoning in the service selection. These issues have
been addressed in [23] which accommodates semantic capability in the context-aware service selection. On the other
hand, works such as [24] and [25] have used users personal
and social information as contexts in the selection method.
The work in [24] makes use of a users social network and
considers the direction of information flowing from the user
device to the social network, users contacts and locations as
the context information. By contrast, work in [25], integrates

3058

IEEE TRANSACTIONS ON PARALLEL AND DISTRIBUTED SYSTEMS,

VOL. 25,

NO. 12,

DECEMBER 2014

Fig. 1. Network model.

the social information in the user devices future locations in

order to select a service provider.
Nevertheless, none of the above mentioned methods
focus on the requested service type while selecting a service.
Therefore, they fail to meet a users service expectations for
different classes of services. The proposed selection methods are tailored for specific service classes such that a users
expectations for that class can be met. It is to be noted that
class of a requested service as well as a users service expectations for that class are defined by user contexts. Besides of
using a number of existing route specific metric and service
specific metric in the proposed methods, this work also
introduces a new service specific metric, quantified user
opinions, for comfort service selection.

PRELIMINARIES

The WLAN, a popular choice in many public places and

industrial environments, is considered in this work as
illustrated in Fig. 1a. The M2M network, considered to be
useful in emergency search and rescue operations, is realized over the WLAN in the proposed selection middleware [15]. The realization of common M2M components
through WLAN components are as follows: the WLAN
central server acts as the M2M server which provisions
safety-related services and the Internet, the access points
serve as the M2M gateways to the mobile users in the
WLAN area which is also referred to as the M2M-area
Network (M2MAN). There are two exit doors in the
M2MAN for evacuation during emergencies.

3.1 Pre-Emergency Environment

Now, Fig. 1a represents the pre-emergency network at a
time T < Te where Te is the time of occurrence of an emergency. At this phase, all users are able to access services
through the M2M server via the M2M-gateways. Besides,
these users are considered to follow the random waypoint
(RWP) mobility model with a walking velocity. It is to be
noted that the RWP mobility model is the most widely used
mobility model in wireless networks. Being simple to model
and simulate, it has gained such an enormous popularity.
Consequently, by considering this mobility model, the proposed pre-emergency WLAN-based M2M network is able
to describe a common and generic network model present
in the current literature.

3.2 Emergency Environments

Next, Fig. 1b illustrates the network immediately after an
emergency when Te < T < Ts with Ts indicating the
time a user have safely evacuated from this affected network.
As explained earlier in Section (3), the WLAN-based M2M
network is comprised of the following essential components:
M2M server, M2M gateways and M2MAN. As the M2M
server is considered to be completely unreachable due to the
infrastructure damages caused by the emergency, the M2M
network fails to provide services to the mobile users in the
M2MAN. These users, however, are still capable to form an
instantaneous network by communicating with other users
within their M2M connectivity range, e.g., Wi-Fi transmission range. Such an infrastructure-less, instantaneous and
ad-hoc network formed by the mobile users is popularly
known as the MANET. Therefore, in absence of the infrastructure-based M2M network, a MANET is assumed to be
formed for service discovery and access after the emergency.
In the MANET, users are aware of other users locations
and velocities within their Wi-Fi transmission ranges, R,
through the regular beacon broadcasts. The mobility of a
user for this duration is likely to be different from the normal
environment. Nevertheless, an accurate mobility model is
required to be considered such that effects of user mobility
on the network topology can be minimized while discovering and selecting services [26]. Hence, we consider that users
follow an emergency mobility model in which they choose
one of the most preferred obstacle-free routes to reach an exit
[27]. Briefly, in this mobility model, users who have witnessed the emergency select one of the available exit doors
resulting the shortest distance from their locations. The other
users obtain movement information of those users through
their M2M connectivity enabled devices and select the exit
which has been selected by the majority of the preceding
users. Note that, to imitate a realistic emergency scenario,
this mobility model considers obstacles in the network and
determines an obstacle-free path using the Voronoi diagram.
Interested readers are referred to [27] for further details.
3.3 Post-Emergency Environments
Then, when the time is T > Ts , Fig. 1c shows that users
have evacuated to a safer network area. Therefore, user
velocity has returned to normal walking velocity and thus
RWP mobility model is considered. In this network, a

SUROBHI AND JAMALIPOUR: A CONTEXT-AWARE M2M-BASED MIDDLEWARE FOR SERVICE SELECTION IN MOBILE AD-HOC NETWORKS

Fig. 2. The proposed service selection middleware.

MANET is formed for discovering service providers as well

as users opinions about the discovered providers in order
to access services. Such opinions can be accessed from
either the local cache of mobile users that stores
previously published opinions on the Web 2:0 applications
or users spontaneous reaction upon receiving requests for
opinion related to a service. Note that the cache and trust
management techniques are not discussed since most of the
existing techniques can be employed in this work.

SERVICE SELECTION MIDDLEWARE

A flow chart of the proposed context-aware M2M-based service selection middleware is presented in Fig. 2 which
explains that a middleware, operating on a users mobile
device, initiates its functions upon receiving a service request
from the user. The service requested is classified using users
physical contexts that represent the network environments,
i.e., the emergency environment is detected from users
velocity which significantly increases during an emergency
for faster evacuation. Hence, from the velocity information,
collected through the Wi-Fi-based M2M connectivity and the
integrated GPS of mobile devices, the requested service is
classified to be either a safety-related or comfort service.
Now, for either class, a MANET is initiated using the computing contexts, i.e., when the dedicated server (or a replica)
is unreachable, the middleware initiates an alternative network connection for discovering the service. The middleware then tailors individual selection methods incorporating
user-related contexts, also referred to as the users service
expectations, for the detected class of services. Such user
expectations for individual service classes are considered to
be pre-defined by the users within the middleware. Details
of accessing the safety-related and comfort services are
described in Sections 4.1 and 4.2 respectively.

4.1 Safety-Related Services

4.1.1 Discovery Phase
For a safety-related service request, when the selection
middleware fails to retrieve any related information from

3059

the locally cached lookup table, it conducts a service discovery phase. This phase is conducted by modifying the
cross-layer service discovery method presented in [18] in
which service is discovered by piggybacking a service
request, SREQ, with the route request, RREQ, of the AODV
routing [28]. The service reply, SREP, from a provider is
then piggybacked with the route reply, RREP, of the
AODV routing. Generally, this routing technique returns
the hop counts between the user and discovered providers,
H1 ; H2 ; . . . ; HK and providers addresses P1 ; P2 ; . . . ; PK for
k 1; 2; . . . ; K providers.
We further modify the aforementioned routing as follows: each providers capacity information C1 ; C2 ; . . . ; CK is
requested with the SREQ which is piggybacked with the
SREP. Besides, each intermediate node is requested with the
RREQ to piggyback their remaining energy E1 ; E2 ; . . . ; EM
and mobility information, i.e., velocity and movement angle
for m 1; 2; . . . ; M intermediate nodes with the RREP. By
using such modifications, a simultaneous routing and service discovery with the intended information for safetyrelated service selection is performed by flooding fewer
messages in the MANET.
The mobility information is readily used to find the link
expiry time, LET, of links present between the user and the
service providers. The LET indicates the duration of time a
user may remain connected with another user (or a provider) before they become disconnected due to mobility
[29]. Now, for two users i and j with location: x1 ; y1 and
x2 ; y2 , velocity: V1 and V2 and phase angle: u1 and u2
respectively, the LET can be calculated as in Eq. 1. Then,
for l links, where l 1; 2; . . . ; L, presented along the route
between the service requesting user, i, and a provider, K,
the link with the shortest LET defines the route expiry time,
RET, as presented in Eq. 2. The RET provides an estimation of the maximum time duration a route would remain
active between the user and the provider [29]. Finally, for
each discovered provider, hop counts, providers capacity,
intermediate users remaining energy and the RET of each
route are stored in the lookup table.
q
pq rs p2 r2 R2  pr  qs2
;
LETij
p2 r2
(1)
where p V1 cos u1  V2 cos u2
r V1 sin u1  V2 sin u2 ; q x1  x2 ; s y1  y2

RETi;K minLET l :

(2)

4.1.2 Selection Phase

The safety-related service selection process is performed if
multiple service providers are returned after the discovery
phase. In order to select one of the most reliable providers
of the requested service, the middleware only considers the
providers meeting the following conditions for selection:
a the providers capacity, C is below a pre-defined capacity threshold, Cmax , which is the maximum capacity of a
provider and b every intermediate node in the route has
energy, E, above an energy threshold, Emin which is the

3060

IEEE TRANSACTIONS ON PARALLEL AND DISTRIBUTED SYSTEMS,

VOL. 25,

NO. 12,

DECEMBER 2014

minimum energy every user must reserve. Therefore, from

the providers satisfying the aforementioned criteria, a provider is selected such that it has the shortest hop count distance from the requesting user and the highest RET. Hence,
for a user, i, with k discovered service providers (where
k 1; 2; . . . ; K), this selection method can be modeled as a
bi-criteria optimization as presented in Eqs. 3 and 4.
This optimization is further subjected to the constraints presented in Eq. 5. In Eqs. 3, 4 and 5, all symbols bear
the same meaning as previously explained.
Jhop minHi;k 

(3)

JRET maxRETi;k 

(4)
Fig. 3. Flowchart of proposed opinion quantification.

s:t:

Ck  Cmax ; Em  Emin :

(5)

This presented bi-criteria optimization problem is of a

conflicting nature. Therefore, it is unlikely to find a solution for this problem that optimizes both objectives
simultaneously. Hence, the popularly used Pareto principle, a concept of Economics which has been widely used
in Engineering to solve conflicting goals, is required to
be implemented [30]. The providers associated with the
Pareto optimal solutions (or weakly Pareto optimal solutions in the case when no Pareto optimal solution exists)
among the discovered providers are therefore can be
considered as the reliable providers. The detailed
description of determining these providers using the Pareto principle is presented in the supplementary section,
which can be found on the Computer Society Digital
Library at http://doi.ieeecomputersociety.org/10.1109/
TPDS.2014.2307875.
Now, any of the discovered providers associated with
the Pareto/weakly Pareto optimal solutions can be selected
as one of the most relaible providers for service access.
Therefore, the user unicasts a service registration request
(SRREQ) message to the selected provider to express the
interest to access the desired service. If no SRREP is received
from the selected provider, the middleware unicasts the
SRREQ messages to the remaining providers sequentially.
The middleware finishes its operation when an SRREP message is obtained or no SRREP message is obtained at all
from any of the providers.

4.2 Comfort Services

4.2.1 Service and Opinion Discovery
If no related provider is found for a comfort-service request
in the lookup table, providers are discovered by following
the AODV-based cross-layer service discovery as presented
in [18] and thus Hk and Pk are stored in the table as previously mentioned. Now, for each discovered provider, the
middleware discovers opinions about them from the 3-hop
neighbors only such that the opinion query and reply traffic
can be localized instead of globally propagating the information. Besides, the additional time required to request and
collect the opinions can also be reduced by limiting the

opinion query only to 3-hop neighbors. Hence, the middleware sends an opinion query, QO OREQ; S; Pk and the
neighbors send the opinion reply (if they have an opinion)
as RO OREPk ; S; Pk where OREQ is the opinion request
for a desired service S from the provider Pk and OREPk are
the number of opinions obtained for the provider. These
opinions are also stored in the lookup table. However, to be
used as the selection criterion, these opinions are required
to be quantified first.

4.2.2 Opinion Quantification

A flow chart of the opinion quantification procedure is
presented in Fig. 3. From the presented flowchart, it can
be explained that the opinions collected for each discovered providers are first pre-processed using several standard pre-processing techniques. Then, they are
vectorized and weighted using the term frequencyinverse document frequency method. Next, the distance
among these opinions are computed in terms of cosine
similarity such that they can be clustered into two clusters using the K-means clustering technique [31]. A cluster head is then selected from both of these clusters to
determine whether they contain positive opinions or
negative opinions. The polarity of a cluster head opinion
and its quantified value are obtained by using the Sentiwordnet scores which provide scores to English words
and their synonyms presented in the lexical database
Wordnet [32], [33]. Once the cluster head opinions are
quantified, the final opinion quantification score for a
provider is computed by combining the quantitative
scores obtained for the clusters and normalizing it to the
total number of opinions discovered for that provider. A
further discussion on the opinion quantification can be
found in the supplementary section, available online.
4.2.3 Service Selection
After the middleware quantifies opinions collected for all
discovered providers, it uses both hop counts information
and the quantified opinion score in a selection function fk as
defined in Eq. 9 which is based on a simple additive weight
function [34]. In Eq. 9, b and 1  b present the weight of

SUROBHI AND JAMALIPOUR: A CONTEXT-AWARE M2M-BASED MIDDLEWARE FOR SERVICE SELECTION IN MOBILE AD-HOC NETWORKS

the hop counts and the quantified opinion score respectively.

fk bHi;k 1  byk :

(6)

Now, the weights are within 01 and specified by the

users which allow them to rank their preferred criterion for
service selection. The provider resulting in the highest fk
value is selected for service access. The service registration
process is then initiated by the middleware which is similar
to the safety-related service registration procedure.

RESULTS AND DISCUSSIONS

5.1 Simulation Environment

For the simulation, a flat square space with area of
200  200 m2 containing 100 mobile users is considered.
The users are allowed to move in this pre-emergency network for at least 120 s before any emergency or service
request is simulated. These users are simulated to follow
the RWP mobility model with the velocity and pause time
within 11:6 ms1 and 01 s respectively to imitate the
slowly walking users with small breaks.
An emergency is simulated at a random location of the
network after 120 s of the simulation with an influence range
of 100 m. To simulate the emergency mobility model, the
user velocity is kept within 26 ms1 . The number of simulated obstacles is between 1020. These obstacles are simulated to be made of different materials and thus a fixed value
corresponding to each material is included in the simulation
to present the signal attenuation [27]. Therefore, a signal
traveling through obstacles is only considered to be received
if the attenuated signal remains over a signal-to-noise ratio
threshold. Now, sensing increased velocity after the emergency, the services requested (simulated only after the emergency) are treated as safety-related services. Thus, the
relevant service discovery and selection are implemented.
User velocity returns to normal walking velocity when they
reach one of the exit doors. From the occurrence of the emergency, the emergency mobility is run for 1;800 s. The simulation terminates earlier if all users have already escaped from
the affected network.
Then, we simulate another flat square space of the same
area as previously mentioned with the total number of evacuated users. The simulation starts (we also reset the simulation clock) with an initial random distribution of user
locations and simulates the RWP mobility with the aforementioned velocity and pause time. Observing normal
walking velocity, all received service requests are treated to
be comfort services. Hence, associated service discovery
and selection are performed. For opinions, a database containing upto 500 real-life opinions extracted from Web 2:0
applications (e.g., Twitter and Foursquare) related to each
service is constructed. During the opinion discovery, the
probability of an intermediate user sharing opinions is random and within 0:21. Then, for the opinion pre-processing, we used the Porter stemming algorithm, word length
and frequency range 330 and 11 respectively. The
clustering terminates if either reaches 50 iterations or the
change in each cluster center is less than 0:001 between iterations. Besides, the b in the comfort service selection function is kept 0:5. This simulation is also run for 1;800 s.

3061

Note that, for all scenarios, service requests are simulated

following the Poisson distribution. The number of simulated service providers (static) is 10 and each provider offers
10 different services. Additionally, the transmission range,
R, of each user is set within 1020 m because they are considered to use Wi-Fi connectivity. Moreover, it is to be noted
that due to the unavailability of the real-life data on the
emergency-affected and post-emergency environments, i.e.,
data describing user movements and service requests, this
work evaluates performances of the proposed middleware
through simulations only. Nevertheless, the simulation
parameters and their values used in this work have been
selected realistically. For instance, practical Wi-Fi transmission range, changes in user movement patterns due to the
occurrence of the emergency, real-life human walking
velocity range for pre-emergency as well as post-emergency
environments, real-life human running velocity range during the emergency, radio signal propagation considering
arbitrarily complex-shaped obstacles during the emergency
(also consequent signal attenuation) and limited energy and
capacity level of the user device to represent the real-life
mobile device characteristics. As a result of the above realistic considerations, the obtained simulation results are
expected to be capable of predicting real-life movements
and data. Nevertheless, the performance of the K-means
clustering technique is evaluated collecting the real-life
opinions, as explained earlier, expressed about invoked
comfort services.

5.2 Performance Metrics

We define the following metrics in order to evaluate the
proposed middlewares performance:
Success rate. The success of service registration for service
NS
100 percent, in
access is defined as the success rate, N
T
which NS is the number of successful registrations and NT
is the total number of attempts made for registration. The
ideal value of the success rate is 100 percent and is practically expected to be closer to it.
Response time. The time between a SREP and SRREP
(4.1.2) received by a user is defined as the response time,
TSREP  TSRREP . A lower delay is expected to access services, especially for the safety-related services, since the
underlying network topology changes frequently resulting
in route failures.
User satisfaction rate, USR. The rate to express the number
of instances a user successfully registers with the best provider (using proposed selection methods) to the total successful registration is defined as the USR. The ideal value of
USR is 100 percent which indicates that a user is always
able to access services from the best provider.
Additionally, the following metrics are defined to evaluate the K-means clustering algorithms performance:
Precision, recall, cluster distance. Given a desired correct
cluster i for some elements and an obtained resultant cluster
j for the same elements, precision and recall can be defined
n
n
as pi;j ni;jj and ri;j ni;ji respectively in which ni;j , ni and nj
are the number of elements from cluster i assigned in the
obtained cluster j, the total number of elements in cluster i
and cluster j respectively. Finally, the cluster distance,
Cd c1  c2 with c1 and c2 being cluster centers of

3062

IEEE TRANSACTIONS ON PARALLEL AND DISTRIBUTED SYSTEMS,

VOL. 25,

NO. 12,

DECEMBER 2014

Fig. 4. The impact of mobile users on success rate.

Fig. 5. The effect of mobile users on response time.

clusters 1 and 2 respectively. The best and worst possible

value of these three metrics are 1 and 0 respectively.

Then, in Fig. 4b, we compare the comfort service selection method with the MHCC method. The proposed method
has a higher success rate than the MHC method because
hop information is only one of the selection criterion.
Besides, the number of collected opinions about providers
of a particular service can be different for two neighboring
users which may result in different ranking of the same providers for those users. This, in turn, further reduces traffic
congestion resulting in a higher success rate. Note that the
transmission range has an effect on the success rate of the
comfort service selection and the MHCC methods similar to
the safety-related service selection and the MHCE methods
for reasons as explained earlier.
Next, we observe from Figs. 5a and 5b respectively that
the safety-related and comfort service selection method, in
general, has a lower response time with fewer users present
in the network. It can be understood that with fewer users
in the network, the amount of collected information is
smaller, i.e., velocity information to compute RET for the
safety-related service selection method and the number of
service-related opinions for the comfort service selection
method. Therefore, shorter time is required to process them
for analyzing the respective selection criteria and ranking
the providers before sending a SRREQ. Conversely, the
amount of collected information increases with an
increased number of users in the network which in turn
increases the time needed for the selection process. Subsequently, a rise in the response time is observed for both of
the proposed method.
To elaborate, we also observe from Fig. 5a that the
MHCE method has a shorter response time for fewer
number of users since the amount of information needs to
be analyzed for service selection is small. Now, with
fewer discovered providers, due to fewer users in the network, the proposed method may not find routes with
high enough RET and thus be required to send SRREQ to
more than one provider before being successful to receive
an SRREP. Hence, it has a response time similar or slightly
higher than the MHCE method for fewer users in the network. By contrast, as the number of users increases, introducing more discovered providers and routes to them, the
probability of obtaining many stable routes to a provider
becomes higher. Hence, the proposed method results in a

5.3 Performance Evaluation

To compare the performances of the proposed selection
methods, the performance of the MHC method, within the
proposed middleware, is obtained during the emergency to
select safety-related services, noted by MHCE , and after the
evacuation to select comfort services, noted by MHCC .
Firstly, we present the impact of the number of mobile users
on the success rate of the selection methods tailored within
the proposed middleware for safety-related and comfort
services in Figs. 4a and 4b respectively. From both figures, it
can be observed that an increase in the number of mobile
users increases the rate of success for all selection methods.
The increased number of users, in general, increases the
probability of discovering more providers, more routes and
thus boosts the success rate.
Now, from Fig. 4a, it can be observed that the proposed
safety-related service selection method outperforms the
MHCE method. In principle, the MHC method allows only
one route (the shortest) to the provider. This route can fail
due to the mobility and can often easily become congested
due to the neighboring users also trying to reach that provider using the same route. In an emergency environment,
with the increased user velocity, the routes become more
unstable than usual which results in the lower success
rate. The proposed method, however, selects providers
with stable routes and thus achieves a higher success rate.
Nevertheless, the high mobility and interference caused
by obstacles restrict the proposed method to achieve
the ideal success rate especially when the transmission
range is short. Hence, the success rate is observed to further improve with an increased transmission range
because it increases the probability of discovering more
providers and routes leading to them. As a result, more
stable routes to providers can be selected mitigating the
effects of mobility and interference. The increased transmission range also increases the success rate of the MHCE
method for similar reasons. Nevertheless, the success rate
of MHCE still remains lower than that of the proposed
safety-related service selection method.

SUROBHI AND JAMALIPOUR: A CONTEXT-AWARE M2M-BASED MIDDLEWARE FOR SERVICE SELECTION IN MOBILE AD-HOC NETWORKS

3063

TABLE 1
User Satisfaction Rate at Different Transmission Ranges

Fig. 6. Evaluation of the impact of network area.

lower response time than the MHCE method. Besides,

from Fig. 5b, it can be explained that due to the time
required to quantify the service related opinions about a
provider the response time of the comfort service selection
method is higher than the response time of the MHCC
method. Both methods, however, face increased response
time with the increased number of users in the network.
Note that both Figs. 5a and 5b show that an increase in
the transmission range increases the response time by
allowing more providers, routes and related opinions to
be discovered by the users. However, characteristics of
the response time of the proposed methods compared to
the corresponding MHC methods remain similar.
Later, in Fig. 6, we present the effect of the network area
on the success rate and the response time for the proposed
selection methods. For both methods, the success rate falls
and the response time rises with an increase in the network area. In fact, a smaller network area enables the users
and providers to be close which in turn results in a higher
success rate and a lower response time for the selection
methods. By contrast, as the network area grows, the distance between the users and providers also increases.
Hence, the possibility of route failure and inability to
discover sufficient providers to make a successful service
selection for a given transmission range become higher. As
a result, the success rate shows a decline and the response
time shows an incline for larger network areas. When compared with the relevant MHC methods, the proposed
methods are observed to follow similar characteristics as
earlier observed.
Furthermore, in Table 1, we present two different
cases (case-1: 80 users, 10 providers, opinion sharing
probability 0:5 and case-2: 50 users, 6 providers, opinion
sharing probability 0:8) of computed USR. The USRs of
the proposed comfort and safety-related service selection
methods are denoted by USR1 and USR2 respectively.
From the results, we can remark that in most cases, for
both methods, a user is able to invoke a service from the
best provider. For a lower transmission range, the USR
downgrades below 80 percent since SRREQ or SRREP
may not be forwarded due to route failures.
Lastly, we present the clustering performance by evaluating p, r and Cd c1  c2 as presented in Fig. 7. The results are

obtained by simulating five groups of opinions ranging

from 10200. It can be observed that the obtained results
are within 0:81 and often close to the ideal value 1. The
clustering performance plays a significant role in determining the opinion scores because an accurate clustering of positive and negative opinions would result in the actual
quantified value of opinions for the providers.

CONCLUSIONS AND FUTURE WORK

We proposed a context-aware M2M-based middleware for

service selection in MANETs. In order to allow access to
services during and after an emergency, the middleware
utilized context information collected through M2M communication between the mobile devices such that at first, a
service request was classified into either safety-related or
comfort services. Then, a MANET was initiated by the middleware for service discovery if the M2M server/replica
M2M server was inaccessible. Finally, to access the service,
the middleware, implemented a selection method proposed
particularly for the detected class of services combining the
users service expectations for that class. Based on several
metrics, simulation results were presented to support the
proposed middleware.
The current research presented a new understanding
of the service selection in MANETs. Nevertheless, this
research considered the RWP mobility model in the preemergency network which is unable to represent human
mobility. Hence, in future, mobility models such as Truncated Levy Walk model that closely represent human
mobility can be considered in the pre-emergency network
to evaluate the network performances. Additionally, this
work assumed that there is no malicious opinion being
shared while collecting opinions for the selection of

Fig. 7. Evaluation of the K-means clustering algorithm.

3064

IEEE TRANSACTIONS ON PARALLEL AND DISTRIBUTED SYSTEMS,

comfort services in the post-emergency networks. Therefore, in the future, security and trust issues during the
opinion collection can also be considered to evaluate the
network performances.

REFERENCES
[1]

[2]

[3]
[4]

[5]
[6]

[7]

[8]

[9]

[10]

[11]
[12]
[13]

[14]
[15]
[16]

[17]
[18]
[19]

G. Paravati, C. Celozzi, A. Sanna, and F. Lamberti, A FeedbackBased Control Technique for Interactive Live Streaming Systems
to Mobile Devices, IEEE Trans. Consumer Electronics, vol. 56,
no. 1, pp. 190-197, Feb. 2010.
A.J. Mashhadi, S.B. Mokhtar, and L. Capra, Habit: Leveraging
Human Mobility and Social Network for Efficient Content Dissemination in Delay Tolerant Networks, Proc. IEEE Intl Symp. A
World of Wireless, Mobile and Multimedia Networks and Workshops
(WoWMoM 09), pp. 1-6, 2009.
J.C. Oberg, A.G. Whitt, and R.M. Mills, Disasters Will HappenAre You Ready? IEEE Comm. Magazine, vol. 49, no. 1, pp. 36-42,
Jan. 2011.
L. Baochun and K.H. Wang, NonStop: Continuous Multimedia
Streaming in Wireless Ad Hoc Networks with Node Mobility,
IEEE J. Selected Areas in Comm., vol. 21, no. 10, pp. 1627-1641, Dec.
2003.
A.N. Mian, R. Baldoni, and R. Beraldi, A Survey of Service Discovery Protocols in Multihop Mobile Ad Hoc Networks, IEEE
Pervasive Computing, vol. 8, no. 1, pp. 66-74, Jan.-Mar. 2009.
C.N. Ververidis and G.C. Polyzos, Service Discovery for Mobile
Ad Hoc Networks: A Survey of Issues and Techniques, IEEE
Comm. Surveys & Tutorials, vol. 10, no. 3, pp. 30-45, Third Quarter
2008.
C. Frank and H. Karl, Consistency Challenges of Service Discovery in Mobile Ad Hoc Networks, Proc. ACM Intl Symp.
Modeling, Analysis and Simulation of Wireless and Mobile Systems
(MSWiM 04), pp. 105-114, 2004.
A. Varshavsky, B. Reid, and E. De Lara, A Cross-Layer Approach
to Service Discovery and Selection in MANETs, Proc. IEEE Intl
Conf. Mobile Ad Hoc and Sensor Systems Conf. (MASS 05), pp. 466474, 2005.
S.E. Athanaileas, C.N. Ververidis, and G.C. Polyzos, Optimized
Service Selection for MANETs Using an AODV-Based Service Discovery Protocol, Proc. Ann. Mediterranean Ad Hoc Networking
Workshop (MEDHOCNET 07), 2007.
S.S. Yau and F. Karim, An Energy-Efficient Object Discovery Protocol for Context-Sensitive Middleware for Ubiquitous
Computing, IEEE Trans. Parallel and Distributed Systems, vol. 14,
no. 11, pp. 1074-1085, Nov. 2003.
V.C. Trinh and A.J. Gonzalez, Discovering Contexts from
Observed Human Performance, IEEE Trans. Human-Machine Systems, vol. 43, no. 4, pp. 359-370, July 2013.
E. Cambria, B. Schuller, Y. Xia, and C. Havasi, New Avenues in
Opinion Mining and Sentiment Analysis, IEEE Intelligent Systems, vol. 28, no. 2, pp. 15-21, Mar./Apr. 2013.
H.K. Kim, J.K. Kim, and Y.U. Ryu, Personalized Recommendation over a Customer Network for Ubiquitous Shopping, IEEE
Trans. Services Computing, vol. 2, no. 2, pp. 140-151, Apr./June
2009.
X. Yang, Y. Guo, and Y. Liu, Bayesian-Inference-Based Recommendation in Online Social Networks, IEEE Trans. Parallel and
Distributed Systems, vol. 24, no. 4, pp. 642-651, Apr. 2013.
G. Wu, S. Talwar, K. Johnsson, N. Himayat, and K.D. Johnson,
M2M: From Mobile to Embedded Internet, IEEE Comm. Magazine, vol. 49, no. 4, pp. 36-43, Apr. 2011.
N.A. Surobhi and A. Jamalipour, Opinion Based Service Selection in a Pervasive Cooperative Consumer Network, Proc. IEEE
Wireless Comm. and Networking Conf. (WCNC 12), pp. 3318-3322,
2012.
V. Lenders, M. May, and B. Plattner, Service Discovery in Mobile
Ad Hoc Networks: A Field Theoretic Approach, J. Pervasive and
Mobile Computing, vol. 1, no. 3, pp. 343-370, 2005.
R. Koodli and C. Perkins, Service Discovery in On Demand Ad
Hoc Networks, I.E.T.F Draft, http://tools.ietf.org/html/draftkoodli-manet-servicediscovery-00, 2002.
P. Makris, D.N. Skoutas, and C. Skianis, A Survey on ContextAware Mobile and Wireless Networking: On Networking and
Computing Environments Integration, IEEE Comm. Surveys &
Tutorials, vol. 15, no. 1, pp. 362-386, First Quarter 2013.

VOL. 25,

NO. 12,

DECEMBER 2014

[20] S. Buchegger and J.Y. Le Boudec, Performance Analysis of the

CONFIDANT Protocol: Cooperation of Nodes-Fairness in
Dynamic Ad-hoc Networks, Proc. ACM MobiHoc 02, pp. 226-236,
2002.
[21] O. Riva and S. Toivonen, A Hybrid Model of Context-Aware Service Provisioning Implemented on Smart Phones, Proc. ACS/
IEEE Intl Conf. Pervasive Services (ISPS 06), pp. 47-56, 2006.
[22] C. Lee and S. Helal, Context Attributes: An Approach to Enable
Context-Awareness for Service Discovery, Proc. IEEE Symp.
Applications and the Internet (SAINT 03), pp. 22-30, 2003.
[23] A.R. El-Sayed and J.P. Black, Semantic-Based Context-Aware
Service Discovery in Pervasive-Computing Environments, Proc.
ACS/IEEE Workshop Services Integration in Pervasive Environments
in IEEE Intl Conf. Pervasive Services (ISPS 06), 2006.
[24] E. Miluzzo, N.D. Lane, S.B. Eisenman, and A.T. Campbell,
Cenceme-Injecting Sensing Presence into Social Networking
Applications, Proc. European Conf. Smart Sensing and Context
(EuroSSC 07), pp. 1-28, 2007.
[25] A. Beach, M. Gartrell, S. Akkala, J. Elston, J. Kelley, K. Nishimoto,
and B Ray, Whozthat? Evolving an Ecosystem for ContextAware Mobile Social Networks, IEEE Network, vol. 22, no. 4,
pp. 50-55, July/Aug. 2008.
[26] H. Nishiyama, T. Ngo, N. Ansari, and N. Kato, On Minimizing
the Impact of Mobility on Topology Control in Mobile Ad Hoc
Networks, IEEE Trans. Wireless Comm., vol. 11, no. 3, pp. 11581166, Mar. 2012.
[27] N.A. Surobhi and A. Jamalipour, A Modified M2M-based Middleware for Movement Prediction in Emergencies, Proc. IEEE
Wireless Comm. and Networking Conf. (WCNC 13), pp. 4599-4604,
2013.
[28] C. Perkins, E.B. Royer, and S. Das, Ad Hoc On-Demand Distance
Vector (AODV) Routing, I.E.T.F Draft, http://www.ietf.org/rfc/
rfc3561.txt, 2003.
[29] W. Su, S.J. Lee, and M. Gerla, Mobility Prediction and Routing in
Ad Hoc Wireless Networks, Intl J. Networking Management,
vol. 11, no. 1, pp. 3-30, 2001.
[30] M. Ehrgott, Multi-Criteria Optimization. Springer-Verlag, 2005.
[31] C. Manning, P. Raghavan, and H. Schtze, Introduction to Information Retrieval. Cambridge Univ. Press, 2008.
[32] S. Baccianella, A. Esuli, and F. Sebastiani, SentiWordNet 3.0:
An Enhanced Lexical Resource for Sentiment Analysis and
Opinion Mining, Proc. Conf. Intl Language Resources and Evaluation (LREC 10), 2010.
[33] About WordNet Princeton Univ. 2010 http://wordnet.
princeton.edu.
[34] C. Hwang and K. Yoon, Multi-Attribute Decision Making. SpringerVerlag, 1981.

SUROBHI AND JAMALIPOUR: A CONTEXT-AWARE M2M-BASED MIDDLEWARE FOR SERVICE SELECTION IN MOBILE AD-HOC NETWORKS

Nusrat Ahmed Surobhi is currently working

toward the PhD degree at the School of Electrical and Information Engineering, University
of Sydney, Sydney, Australia. Her research
interest includes machine-to-machine communications and user-generated traffic in wireless
networks. She has served as a reviewer for
several flagship IEEE conferences and journals. She has received several scholarships
including Endeavour International Postgraduate Research Scholarship funded by the Australian Government. She has been a student member of the Institute
of Electrical and Electronics Engineers (IEEE) since 2007.

3065

Abbas Jamalipour (S86-M91-SM00-F07)

received the PhD degree in electrical engineering
from Nagoya University, Japan. He is the professor of Ubiquitous Mobile Networking at the University of Sydney, Australia. He is the author of six
technical books, nine book chapters, more than
300 technical papers, and three patents, all in the
area of wireless communications. He was the editor-in-chief of IEEE Wireless Communications
(2006-08), vice president-conferences (2012-13)
and a member of Board of Governors of the IEEE
Communications Society, and has been an editor for several journals.
Previously he has held positions of the chair of the Communication
Switching and Routing and the Satellite and Space Communications
Technical Committees and the vice director of the Asia Pacific Board, in
ComSoc. He was a general chair or Technical Program chair for a number
of conferences, including IEEE ICC, GLOBECOM, WCNC and PIMRC.
He is also an elected member of the Board of Governors (2014-16), IEEE
Vehicular Technology Society. He has received a number of prestigious
awards such as the 2010 IEEE ComSoc Harold Sobol Award, the 2006
IEEE ComSoc Distinguished Contribution to Satellite Communications
Award, the 2006 IEEE ComSoc Best Tutorial Paper Award. He is a fellow
of the Institute of Electrical, Information, and Communication Engineers
(IEICE) and the Institution of Engineers Australia, an ACM professional
member,and an IEEE distinguished lecturer. He is a fellow of the IEEE.

" For more information on this or any other computing topic,

please visit our Digital Library at www.computer.org/publications/dlib.