Towards a Semantic and Dynamic Cluster based Web Service Discovery

System for Ubiquitous Environments

Rim Helali

, Nadia Ben Azzouna

and Khaled Ghedira

Higher Institute of Management, Research Laboratory SOIE, Tunis University, Tunis, Tunisia

National School of Computer Science, Research Laboratory SOIE, Manouba University, Manouba, Tunisia

Keywords:

Ubiquitous Environments, Web Service Discovery, Clustered Architecture, Semantic Service Publication and

Discovery.

Abstract:

Ubiquitous computing aims to exchange and share services anywhere, anytime. However, discovering the

appropriate service in ubiquitous environments poses very speciﬁc challenges. In this paper, we propose a

new Semantic and Dynamic Cluster based Web Service Discovery System (SDC- WSDS) which deals with

dynamicity and scalability challenges. Our discovery system is based on a clustered architecture and supports

semantic publication and discovery queries. We, hence, present a new clustering algorithm for ubiquitous

environments inspired by the Weighted Clustering Algorithm (WCA) and a semantic service publication and

discovery model. This model is based on attribute value pairs and a semantic distance function.

1 INTRODUCTION

”Ubiquitous computing is a technology in which in-

visible computers are embedded and connected with

all things so that anyone can communicate, exchange

and share information anywhere, anytime” (Kang

et al., 2007). Currently, we are surrounded by com-

puters, mobile phones and personal digital assistants

(PDA). It becomes difﬁcult to manage these devices

heterogeneity, conﬁgure different types of applica-

tions and dynamically discover appropriate services.

Web service technologies seem to be adequate to ad-

dress the heterogeneity problem and to help imple-

menting pervasive applications. It is due to the fact

that a web service is like an abstract middleware hid-

ing the details about the application platform and pro-

gramming language (Xu et al., 2008). In addition to

devices heterogeneity, many other challenges related

to ubiquitous environments were addressed in several

articles (Estrin et al., 2002) (Satyanarayanan, 2001).

In fact, while discovering web services we are faced

to very speciﬁc ones. We focus in this paper mainly

on those that we consider the most critical for web

services discovery such as:

• Dynamicity: refers to the system’s ability to man-

age the mobility of devices in the environment

and their unexpected appearance or disappear-

ance. Due to providers and directory nodes dy-

namicity (nodes that maintain descriptions of web

services), the system must be able to replace them

at any time (Abdennadher, 2008). Users mobility

must be also taken into account in order to ensure

web services availability wherever users move.

• Scalability: refers to the system’s ability to main-

tain good performance with the increased num-

ber of service publication and discovery queries

(Gao and Steenkiste, 2004). In fact, the majority

of ubiquitous environments are characterized by

the abundance of users and their interactions, and

by the big number of connected devices and ser-

vices. To ensure scalability, centralized solutions

are to be avoided in order to prevent the problems

of bottleneck and Single Point Of Failure (SPOF).

Priority should also be given to localized interac-

tions; we refer here to localized scalability mean-

ing that remote queries should be eliminated. In

fact, the density of interactions should decrease as

soon as the user moves away; otherwise the net-

work will be overwhelmed by costly interactions

that are of little interest (Satyanarayanan, 2001).

In this paper, we present our Semantic and Dynamic

Cluster based Web Service Discovery System (SDC-

WSDS) which is based on a clustered architecture and

supports semantic publication and discovery queries.

Because of the abundant number of users charac-

terizing the majority of ubiquitous environments, a

clustered architecture is adopted to reduce the cost

295

Helali R., Ben Azzouna N. and Ghedira K..

Towards a Semantic and Dynamic Cluster based Web Service Discovery System for Ubiquitous Environments.

DOI: 10.5220/0003972902950300

In Proceedings of the 14th International Conference on Enterprise Information Systems (ICEIS-2012), pages 295-300

ISBN: 978-989-8565-11-2

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

of searching services. Services are mapped using a

semantic distance function in order to improve the

service searching efﬁciency. Dynamicity of nodes,

and especially directory nodes, is managed by ap-

plying a dynamic clustering algorithm inspired from

the Weighted Clustering Algorithm (WCA) origi-

nally proposed for ad hoc mobile networks (Chatter-

jee et al., 2001). We adapted the WCA algorithm

to cope with the speciﬁcities of ubiquitous environ-

ments. Hence, we aim in this work to deﬁne a system

that manages the high dynamicity of the environment

while guaranteeing the scalability requirement.

The rest of the paper is organized as follows. Sec-

tion 2 addresses related work and identiﬁes some lim-

itations. Section 3 describes the SDC-WSD system.

Finally, Section 4 concludes the paper and discusses

some future directions.

2 RELATED WORK

Web service discovery systems proposed in the liter-

ature used a diversity of architectures ranging from

centralized systems to fully distributed systems. De-

spite the fact that availability and rapidity of access

are guaranteed with systems using centralized direc-

tories (Chen et al., 2006), (Hwang et al., 2007), these

systems suffer from problems like SPOF and bot-

tleneck as the number of registration and discovery

queries increases. To address the scalability issues

related to centralized architectures, solutions have

been proposed dealing mainly with decentralization

of directory nodes (Kang et al., 2007), (Xu et al.,

2008), (Gao and Steenkiste, 2004), (Kim et al., 2005),

(Liu et al., 2002). Kim et al. proposed a fully

distributed architecture by electing for every service

requestor or provider, K volunteer nodes (directory

nodes) to which publication and discovery queries are

sent (Kim et al., 2005). Problems of SPOF and bottle-

neck are thus avoided but this decentralization has its

cost because voluntary nodes are announced by ﬂood-

ing the query and even in case of updating the volun-

teers list, ﬂooding is used. To avoid the ﬂooding prob-

lem, Xu et al. and Kang et al. proposed web service

discovery systems based on a clustered architecture

that reduces the search space and thus avoids over-

loading the network (Kang et al., 2007), (Xu et al.,

2008). To add semantic search capability, Kang et

al. proposed a clustering method based on seman-

tic similarity where each cluster contains descriptions

of the most semantically similar services. Services

are in fact, deﬁned by attribute-value pairs (AV- Pairs)

and semantics of each attribute is described using an

ontology (Kang et al., 2007). In order to manage

nodes dynamicity characterizing ubiquitous environ-

ments, many solutions have been proposed ranging

from choosing nodes that have the highest resources

capacity and the lowest mobility to be directory server

nodes (Kim et al., 2005), to proposing a cluster forma-

tion algorithm inspired from the cluster head selection

method (Liu et al., 2002).

In our web service discovery system SDC-WSDS,

we propose 1) a clustering algorithm inspired from

the Weighted Clustering Algorithm (WCA) originally

proposed for ad hoc mobile networks and which we

adapted to meet with ubiquitous environments chal-

lenges and 2) a semantic service publication and dis-

covery model based on AV-Pairs and a semantic dis-

tance function.

3 SDC-WSDS DESCRIPTION

In this section, we present our web service discov-

ery system SDC-WSDS which is designed to en-

hance scalability, and dynamicity while discovering

web services in ubiquitous environments. To provide

scalable web service discovery, we adopt a clustered

architecture in order to reduce the number of publi-

cation and discovery queries (the access is limited to

the directory nodes). Each cluster is organized in a

two level hierarchical architecture containing a Direc-

tory Server (DS) and a number of web service servers

(WSS) (Xu et al., 2008). Thus, the ubiquitous envi-

ronment is partitioned into dynamic clusters using a

clustering algorithm as detailed in section 3.1. In our

proposed architecture, as described in ﬁgure 1, each

node can be mobile and can take one of four roles:

• A Directory Server (DS): is responsible of a cer-

tain region in the ubiquitous environment. It

maintains descriptions of all existing web services

distributed within its region (Xu et al., 2008). Dif-

ferent from those in USDM-PerCom (Xu et al.,

2008), each DS contains, in addition to the de-

scriptions of web services within its region, de-

scriptions of the most semantically similar web

services according to the AV-Pairs assigned to

each DS. In fact, in order to be able to cal-

culate the semantic distance, AV-Pairs (provided

from the ﬁrst service description that a local WSS

sends) are assigned to DSs. After registering its

ﬁrst service, each DS sends the AV-Pairs, that it

will be responsible of, to all other DSs. The DS

having not yet been assigned any values, when re-

ceiving a new service description, checks the val-

ues supported by the other DSs by calculating a

semantic distance between them and the received

values. If the calculated distance is lower than a

ICEIS2012-14thInternationalConferenceonEnterpriseInformationSystems

296

Figure 1: SDC-WSDS architecture.

predeﬁned threshold, the DS registers the service

description without setting it as its supported val-

ues and waits for another service description.

• A backup Directory Server (backup DS): main-

tains the same service descriptions as the local

DS (periodic messages are sent from the DS to

the backup DS containing a list of the new regis-

tered services). It is used to replace the DS in case

of failure in order to guarantee service availability

and to enhance stability of the discovery system.

• A Web Service Server (WSS): contains a number

of web services and their descriptions. A WSS is

assigned to a DS in the ubiquitous environment.

This DS becomes its gate to the entire ubiquitous

environment (Xu et al., 2008).

• A service requester: accesses to the environment

through a portal and speciﬁes the characteristics

of the desired service (Xu et al., 2008). The portal

is responsible of articulating the request into AV-

Pairs and sends the query to the nearest DS .

To enhance the effective search capability, the most

relevant service according to the query has to be dis-

covered without visiting all the DSs in the environ-

ment. Thus, publication or discovery queries are only

sent to the most semantically similar DSs. A seman-

tic distance function is used to calculate the similarity

between registered services and publication and dis-

covery queries. The following subsections describe

in details the proposed solutions to realize our SDC-

WSDS.

3.1 Clustering Algorithm

To form the clusters of the environment, we deﬁned

a clustering algorithm inspired from the WCA al-

gorithm. The choice of WCA is based on the fact

that this algorithm takes into consideration several pa-

rameters like transmission power, mobility and bat-

tery power of mobile nodes (Chatterjee et al., 2001).

These parameters are very signiﬁcant for the election

of the cluster-heads (DSs in our case), especially in

ubiquitous environments where nodes are character-

ized by a high dynamicity.

3.1.1 DSs Election and Clusters Formation

Initially, there are no DSs in the environment. The

DSs list is generated for the ﬁrst time by invoking the

DSs election procedure at the time of the system ac-

tivation. Each node in the environment broadcasts to

its immediate neighbors (i.e. one-hop neighbors) its

ID and chooses only the nodes with a transmission

delay lower than a predeﬁned threshold to form its

neighbors set. This restriction on the neighbors set is

mainly motivated by reducing the distance between a

node and its neighbors. The procedure, as described

in Algorithm 1, consists of electing the DSs on the

base of a calculated value W v. In the original WCA

algorithm, the value W v calculated for every node is

a combination of components with certain weighting

factors chosen according to the system needs (Chat-

terjee et al., 2001). Due to the dynamicity of the

ubiquitous environment and in order to guarantee the

system stability, we judge, in our proposed algorithm,

that the nodes mobility and the degree of nodes (i.e.

the number of neighboring nodes dv) are more impor-

tant than battery power in the DSs election procedure.

Therefore, the value W v is calculated as :

W v =

(1 + b∆V + cMv)

(1)

where b and c are calibration factors. The ﬁrst compo-

nent ∆v computes the degree-difference for each node

v by calculating the difference between the ideal num-

ber of nodes a DS can support δ and the number of

neighbors dv. It helps, thus, choosing the node with

the nearest degree to the ideal degree δ as a DS. This

is to ensure that the DSs are not over-loaded and the

efﬁciency of the system is maintained at the expected

level. The component Mv represents the mobility of

nodes. A node with less mobility is always a better

choice to be a DS. To calculate nodes mobility, we

chose, rather than using the mobility metric proposed

in the original algorithm that is based on a localiza-

tion system, to use the mobility metric proposed in (Li

et al., 2007). The choice of this speciﬁc metric is justi-

ﬁed by the fact that it does not depend on any location

system (e.g. GPS) and it can fully capture the rela-

tive motions between a node and its neighborhood, in

real-time, using simple triangulation (Li et al., 2007).

The last component Pv, represents the remaining bat-

TowardsaSemanticandDynamicClusterbasedWebServiceDiscoverySystemforUbiquitousEnvironments

297

Algorithm 1: Clustering algorithm proposed for the SDC-

WSDS.

Input : listID, n, TD, Pv, δ

Initialization : dv=0

for i = 1 to n do

for j = 1 to n − 1 do

if Transmission Delay h TD then

dv = dv + 1

add(listID( j), N)

end if

end for

∆v = |dv-δ|

Mv(t −∆t) =

|U|

∑

i, j∈U

i j

(t−∆t)

∆t

W v =

(1 + b∆V + cMv)

end for

listId = list of the identiﬁers of one-hop neighbors, TD = a predeﬁned transmission delay

threshold that a node cannot exceed to be considered as a neighbor, n = the total number

of one-hop neighbors, N = the set of one-hop neighbors of the node v with transmission

delay h TD, Pv = Battery power at the time t.

tery power of a node. Electing the node which has

the highest battery power is more relevant in the DSs

election than choosing the one which has consumed

the lowest amount of battery power(used in the orig-

inal algorithm). It is due to the fact that a node may

initially have limited battery power. So knowing the

consumed battery power of a node does not reﬂect the

time during which a node can play the role of a DS af-

ter being elected. After calculating its W v, each node

sends its value to its neighbors (nodes forming the

set N). This exchange of messages permits the nodes

to be aware of the node with the lowest W v in their

neighbors set which is elected as a DS. Non elected

nodes (i.e. WSSs) send a membership query to their

local DS. Thus, each new elected DS is aware of its

cluster members. If a node is elected as a DS (i.e.

receives at least one membership query) despite the

fact that it has not the lowest Wv in its neighbors set

(it has the lowest W v in another node neighbors set),

the elected node has to disjoin the cluster to which it

has been assigned as a WSS by sending a disjunction

query to the local DS. Each elected DS chooses from

the cluster the node that has the second lowest W v as

a backup DS. If a node has an empty neighbors set

due to transmission delay restriction, it is then elected

as a DS in its region.

3.1.2 Nodes Dynamicity Management

Due to the dynamic nature of the ubiquitous environ-

ment, the nodes as well as the DSs tend to move in

different directions. So, a node may be detached from

its local DS and looks for a new DS; a reafﬁliation is

thus done. New DSs can be added to the set if a node

cannot ﬁnd a DS to which it can be afﬁliated due to

the overload of neighboring DSs. In addition to nodes

mobility, unavailability of nodes (i.e. breakdown, bat-

tery power extinction) has to be managed especially

for DSs and backup DSs. We discuss each of these

cases in details in Table 1.

3.2 Semantic Discovery Procedure

Publication or discovery of a web service is done by

sending a query containing a description of the ser-

vice. We propose a semantic discovery procedure in

which each query is comprised of a set of AV-pairs

where service functionalities, inputs, outputs, and the

characteristics of the device containing the service are

described by pairs of attributes and values. The choice

of this type of representation is due to the fact that it

allows service providers (WSS) to precisely describe

their services and requesters to easily describe their

service requirements. The semantic of each attribute

is represented by an ontology which provides infor-

mation about the semantic relations among the con-

cepts. The ontology trees are shared by all the nodes

including DSs and WSSs. Thus, nodes are aware of

which type of ontology should be used to describe the

services to be published or discovered (Kang et al.,

2007). A semantic distance function is used to calcu-

late the distance between the AV-pairs of the service

to discover and the AV-Pairs assigned to each DS in

order to enhance the effective search capability. In

fact, each DS is responsible of the most semantically

close service descriptions to its supported AV-Pairs. It

is made possible by calculating the semantic distance

between the AV-pairs of the service to publish and the

AV-Pairs assigned to the DS. Thus, the most appro-

priate service for the considered request is discovered

without visiting all the DSs and therefore we avoid

the network overload.

3.2.1 Web Service Publication

In SDC-WSDS, we register web services not only in

the local DS but also in the K most similar DSs ac-

cording to a calculated value D which is a combina-

tion of the semantic distance and the physical dis-

tance. The duplication of service description in K

DSs is done to ensure availability of services and the

choice of the K DSs on the base of semantic simi-

larity enhances effective search capability (Liu et al.,

2002). First, each WSS which has a new web ser-

vice to publish sends a publication query to its local

DS. The DS calculates the semantic distance between

the service to publish and the AV-Pairs assigned to

the DSs. After calculating the semantic distance, we

choose K DSs to which the service description will

be sent. The parameter K as deﬁned by Liu et al. is

ICEIS2012-14thInternationalConferenceonEnterpriseInformationSystems

298

Table 1: Nodes dynamicity management in SDC-WSDS.

Condition Considered action(s)

A node detecting a new DS closer than its local DS (a

higher signal strength).

The node sends a request for afﬁliation. The DS accepting

the request updates its list and informs the old DS.

Arrival of a new node which sends an afﬁliation request to

all its neighboring DSs but no response is received (due to

overloading of neighboring DSs).

The DS election procedure is invoked, but only neighbor-

ing DSs are involved. After reforming the clusters, each

old DS sends to its local DS its registered service descrip-

tions, and each new DS sends a DS replacement query to

all other DSs in the environment.

Failure of a DS (If no messages are received by the backup

DS during a speciﬁc period of time).

The backup DS initiates the DS election procedure by

sending a DS election query to the WSSs members of the

cluster. After electing the DS, the backup DS sends its reg-

istered service descriptions to the new DS which chooses

the node with the second lowest Wv as its new backup DS

and sends to the other DSs in the environment a message

informing them of its new status.

Failure of a backup DS (if no acknowledgements are re-

ceived during a speciﬁc period of time after sending mes-

sages containing the new registered services).

The DS selects the node that has the smallest Wv from its

WSSs to be the new backup DS (Nodes have to recalculate

their Wv).

Failure of both DS and backup DS (the unavailability of

the DS is detected by the WSSs).

The WSSs have to elect a new DS among them by recal-

culating their Wv and choosing the WSS with the smallest

Wv. The second WSS to have the smallest Wv is chosen

as a backup DS. The newly elected DS broadcasts its infor-

mation to all other DSs. Upon receiving the message, they

add the new DS to their list. If it is semantically and phys-

ically close; they send their registered service descriptions

to it.

equal to log

1−p

A with p is the unavailability prob-

ability of a DS and A is the availability requirement

of a DS (Liu et al., 2002). We propose to calculate a

distance D which value is given by :

D = P

× d + P

× hopcount (2)

where P

and P

are weighting factors with the sum

equal to 1, d is the semantic distance calculated ac-

cording to a semantic distance function such as the

distance function used in (Kang et al., 2007) and hop-

count is the physical distance between the local DS

and each one of the other DSs. This distance D is

used to choose the K DSs for the service publication.

It is a combination of the semantic distance and the

physical distance. The choice is justiﬁed by the fact

that the most semantically close DSs to the service

to publish could be the most distant physically to the

DS. Therefore, a compromise has to be established

between semantic and physical distance in order to

reduce the number of hops when publishing a service,

and according to the provider needs, a bigger impor-

tance can be accorded to one of the two distances by

assigning a larger weighting factor. The K DSs with

the smallest value D are chosen to publish the ser-

vices.

3.2.2 Web Service Discovery

When looking for a service, a service requester ac-

cesses to a portal in which he expresses his request

by specifying the functionalities of the required ser-

vice. The portal takes the responsibility of articu-

lating the request in the form of AV-Pairs and sends

it to the nearest DS. The DS calculates the semantic

distance between the service description and the as-

signed values of all the DSs in the environment and

sends a multicast query to the K ﬁrst DSs with the

smallest semantic distance d. Each DS checks its list

of registered services to see whether there are any ser-

vices matching the query. A DS ﬁnding a matching

service sends a response to the DS originator of the

query with the service ID. After receiving all the re-

sponses, the DS proceeds to a service selection based

on contextual attributes such as the WSS status (busy

or not), the physical distance and so on. A query is

then sent to the DS responsible of the selected service

which sends a service invocation to the WSS contain-

ing the web service. The response of the WSS is then

routed to the service requestor.

4 CONCLUSIONS AND FUTURE

WORK

In this paper, we introduced a semantic and dynamic

cluster based web service discovery system named

SDC-WSDS which is based on a clustered architec-

ture and supports semantic publication and discov-

ery queries. Therefore, we proposed a clustering al-

TowardsaSemanticandDynamicClusterbasedWebServiceDiscoverySystemforUbiquitousEnvironments

299

gorithm permitting the partition of the environment

into dynamic clusters and the management of nodes

dynamicity. Services are semantically deﬁned us-

ing attribute-value pairs and the semantic similarity

between them is measured by a semantic distance

function. Publication and discovery queries are thus

routed only to the most semantically similar nodes in

order to reduce the number of visited nodes and to

enhance searching capability. Our future work will

ﬁrst include the implementation and the evaluation

of SDC-WSDS by testing the stability of the system

in terms of number of nodes reafﬁliations and DSs

set updates and by testing the scalability of SDC-

WSDS in terms of number of publication and discov-

ery queries. Second we propose the extension of the

proposed system to support quality of service during

the web service selection in order to meet with the

invisibility challenge.

REFERENCES

Abdennadher, I. (2008). Une approche pour l’assurance des

qualites de services des systemes publier/souscrire de-

ployes sur un reseau mobile ad-hoc. Master’s thesis,

Ecole Nationale d’Ingnieurs de Sfax, Tunisie.

Chatterjee, M., Das, S. K., and Turgut, D. (2001). Wca: A

weighted clustering algorithm for mobile ad hoc net-

works. Journal of Cluster Computing (Special Issue

on Mobile Ad hoc Networks), 5:193–204.

Chen, I. Y., Yang, S. J., and Zhang, J. (2006). Ubiquitous

provision of context aware web services. IEEE In-

ternational Conference on Services Computing, pages

60–68.

Estrin, D., Culler, D., Pister, K., and Sukhatme, G. (2002).

Connecting the physical world with pervasive net-

works. IEEE Pervasive Computing, 1:59–69.

Gao, J. and Steenkiste, P. (2004). Design and evaluation of

a distributed scalable content discovery system. IEEE

Journal on Selected Areas in Communications, 22:54–

66.

Hwang, Y.-Y., Oh, I.-J., Im, H.-J., Lee, K.-C., Lee, K., and

Lee, S. (2007). Uws broker for ubiquitous web ser-

vices dynamic discovery. International Conference on

Intelligent Pervasive Computing, pages 204–209.

Kang, S., Kim, D., Lee, Y., Hyun, S. J., Lee, D., and Lee,

B. (2007). A semantic service discovery network for

large-scale ubiquitous computing environments. ETRI

Journal, 29:545 – 558.

Kim, M., Kumar, M., and Shirazi, B. (2005). Service dis-

covery using volunteer nodes for pervasive environ-

ments. International Conference on Pervasive Ser-

vices, pages 188–197.

Li, Z., Sun, L., and Ifeachor, E. C. (2007). Gps-free mobil-

ity metrics for mobile ad hoc networks. IET Commu-

nications, (5):1–18.

Liu, J., Zhang, Q., Li, B., Zhu, W., and Zhang, J.

(2002). A uniﬁed framework for resource discov-

ery and qos-aware provider selection in ad hoc net-

works. ACM Mobile Computing and Communications

Review, 6:13–21.

Satyanarayanan, M. (2001). Pervasive computing: Vi-

sion and challenges. IEEE Personal Communications,

8:10–17.

Xu, T., Ye, B., Kubo, M., Shinozaki, A., and Lu, S.

(2008). A gnutella inspired ubiquitous service discov-

ery framework for pervasive computing environment.

8th IEEE International Conference on Computer and

Information Technology, CIT, pages 712 – 717.

ICEIS2012-14thInternationalConferenceonEnterpriseInformationSystems

300