SERVICE ACQUISITION AND VALIDATION
IN A DISTRIBUTED SERVICE DISCOVERY SYSTEM
CONSISTING OF DOMAIN- SPECIFIC SUB-SYSTEMS
Deniz Canturk and Pinar Senkul
METU, Computer Engineering Department, 06531 Ankara, Turkey
Keywords: Distributed, Domain specific and Semantic web service discovery, Ontology, Quality of Service.
Abstract: With the increase in the number of advertised web services, finding the web services appropriate for a given
request will result in problems in terms of performance, efficiency and quality of the discovered services. In
this paper, we introduce a distributed web service discovery system which consists of domain-specific sub-
systems. We establish ontology based domain specific web service discovery sub-systems in order to
discover web services on private sites and in business registries, and in order to improve service discovery
in terms of efficiency and effectiveness. An important problem with effectiveness is to check whether a
discovered service is active or not. Many services in the registries are not active any more. Distributing the
discovery to domain-specific subsystems makes it possible to lower the service discovery duration and to
provide almost up-to-date service status. In this work, we describe the service acquisition and validation
steps of the system in more detail and we present experimental results on real services acquired on the web.
1 INTRODUCTION
Web services begin to play important role in Web
computing. However, finding the most appropriate
Web service among the vast amount of advertised
services becomes a hard task. Different approaches
with specific designs and realization are introduced
to overcome this problem. However these proposed
approaches have some drawbacks.
Most of the Web service discovery approaches
are based on syntactic matching between the service
advertisement and the service query. Popular service
registries such as UDDI (Universal Description,
Discovery and Integration) (uddi, 2004) and ebXML
(Electronic Business using XML) (ebXML, 2005)
provide ways for locating businesses, however, they
are limited to keyword-based search techniques. In
addition, they do not provide QoS information and
service lifecycle tracking since they are
disconnected from the services they advertise.
Most of the proposed systems are not scalable,
i.e. they have difficulty in handling the large number
of service providers and service requesters. This is
basically due to that fact, Web services are collected
in a central registry or central service databases.
When the database becomes overloaded, scalability
problem is resolved by using replica of central
registry of database. However replica management
consumes high operational and management
expenses.
Another important drawback is that current
solutions are incompatible with each other.
Existence of heterogeneous service registries
(UDDI, ebXML), private service portals and
discovery approaches (SPIDeR (Sahin, Gerede,
Agrawal, Abbadi, Ibarra and Su, 2005) or
METEOR-S (Meteor, 2006)) makes it complicated
to publish and search a service. Users spend too
much time to visit numerous user interfaces, to
understand the way to use them, to enter query text
again and again, and assembly the results from each
interface.
Finding the most appropriate Web service for
user’s request manually is a tedious task under these
conditions. Although there are many services being
published, they are scattered on the Web and there is
not a feasible solution yet to bring them together.
This problem is similar to information search issue,
which is handled by search engines (such as Google,
Yahoo, etc.) nevertheless there is no such
established solution for Web service querying.
Therefore, we aim to develop such a solution and
eliminate the burden of facing different web service
93
Canturk D. and Senkul P. (2010).
SERVICE ACQUISITION AND VALIDATION IN A DISTRIBUTED SERVICE DISCOVERY SYSTEM CONSISTING OF DOMAIN- SPECIFIC SUB-
SYSTEMS .
In Proceedings of the 12th International Conference on Enterprise Information Systems - Software Agents and Internet Computing, pages 93-99
DOI: 10.5220/0002900700930099
Copyright
c
 SciTePress

registries, by using a single search interface with
semantic and QoS querying capability. However,
such a system should not have a centralized structure
in order to avoid efficiency and scalability problems.
For this reason, we propose a service discovery
system that uses Domain-Specific Web Service
Discoverer (DSWSD-S) sub-systems which are built
around ontology. Since a (DSWSD-S) has its own
compact ontology, shorter update time and querying
is provided. This structure also supports scalability.
The proposed architecture supports QoS aware
service discovery with syntactic and semantic
matching capabilities while not disturbing the
autonomy of web service discovery. In addition,
QoS information can be collected automatically.
Construction of a DSWDS-S provides a framework
for semantic annotation of available web services, as
well.
Acquisition of the web services is the first step of
a DSWSD-S. In this step, the services that are
related with a given ontology are extracted and
analyzed. It includes the analysis of parameters as
well. In this work, the emphasis is on the acquisition
step of DSWSD-S.
The rest of this paper is organized as follows: In
Section 2, we describe the DSWSD-S architecture.
In Section 3, the design and implementation of
service acquisition and validation in DSWSD-S is
presented. Section 4 contains the experimental
results on the acquisition and validation of real web
services. Related work is given in Section 5. Finally,
conclusion is presented in Section 6.
2 DISTRIBUTED SERVICE
DISCOVERY SYSTEM
CONSISTING OF
DOMAIN- SPECIFIC
SUB-SYSTEMS
2.1 Overall Architecture
The proposed system consists of domain-specific
subsystems, each of which is specialized for an
ontology, that is, each subsystem crawls, discovers
and indexes web services according to its own
ontology. Main idea in the proposed system is to
split the discovery task into subtasks on the basis of
the domains specified with an ontology, in order to
provide parallel processing and scalability. In
addition to these advantages, it facilitates load
balancing in domain-specific web service database
generation and collaboration between DSWSD-Ss in
web service querying. Crawling process continues in
a cycle in order to keep the web service database up-
to-date. As the crawling cycle gets shorter, web
service suggestion becomes more reliable. The
distributed structure supports shorter crawling cycles
by distributing the load to subsystems.
The overall structure is depicted in Figure 1. The
architecture for DSWSD-S consists of two layers:
domain-specific crawler layer and domain-specific
service discovery layer.
2.1.1 Domain-specific Crawler Layer
Web service database generation according to its
own ontology is the main task of a domain-specifi
crawler layer. Web service database generation
process starts with URL and service description
acquisition of advertised Web services. In order to
collect the descriptions, a focused crawling is
performed based the ontology of DSWSD-S. The
next step is to check whether the acquired web
services are still active and accessible or not, by
calling simple yet appropriate input parameters.
Figure 1: Distributed Service Discovery Architecture.
In service extraction phase, validated web
service’s input and output parameters are compared
with the ontology terms in order to determine the
ontological position of the web service. Both
syntactic and semantic matching is used in this
phase. Semantic matching provides better results
than syntactic matching. However semantic
matching requires input and output parameters of the
web service to be annotated by using the DSWSD-S
ontology. Considering that the most of the available
web services do not have semantic annotation,
DSWDS-S supports semantic matching as well.
In service verification phase, extracted services
are verified whether web services actually perform
the advertised tasks. This is a hard and time
consuming process since (i) what a service provides
ICEIS 2010 - 12th International Conference on Enterprise Information Systems
94

should be automatically discovered, (ii) appropriate
inputs should be set, and (iii) it should be
determined that the appropriate outputs are obtained
for the given inputs. If the inputs and outputs are
semantically annotated, service verification can be
fulfilled with the help of ontology. Otherwise, input
and output parameters are compared with the
ontology terms to find the matching degree of
parameter to the DSWSD-S ontology.
In quality of service (QoS) determination step,
web service is evaluated for non-functional
attributes such as response time, throughput,
reliability and availability. In order to obtain QoS
values automatically, the information obtained in
validation phase is recorded and analyzed.
2.1.2 Domain-Specific Service Discovery
Layer
This layer provides a user interface to query the
ontology specific Web service database. The
provided information can be both in syntactic and
semantic form. User can also set QoS constraints for
the desired Web service.
As the first step, keywords provided by the user
are compared with the ontology terms in order to
figure out the ontological position of the required
service. When the keyword matching value is above
the confidence level, its ontological children terms
are also added to search set. If the overall matching
value is above the confidence level, web discovery is
performed in the database of the initially accessed
DSWSD-S. Otherwise, request is considered as
unrelated with this domain and forwarded to other
domain specific web service discovery peers.
DSWSD’s database is queried to discover the
web services fulfilling the request. In addition to
syntactic matching of entered keywords, semantic
keyword sets are used in order to find semantically
related web services. Fast access to database records
is achieved by using inverted index. This table,
which is filled in service extraction phase, includes
ontology term, web service URL and matching
degree columns. In service evaluation phase, web
service suggestion values are determined with
respect to the conformance to request and QoS
values of the web service. Web services are sorted
with suggestion values and best ones are presented
to the user.
2.2 Network Topology among
DSWSD-S Nodes
DSWSD-Ss establish peer-to-peer (P2P) network in
order to communicate with each other. Custom p2p
network infrastructure is implemented instead of
using the p2p network implementations in literature
such as Gnutella (Gnutella, n.d.) or Chord (Morris,
Karger, Kaashoek and Hari Balakrishnan, 2001).
Communication among DSWSD-S is handled by
simple text messaging. In other words, a DSWSD-S
sends XML based text message requests to the other
entire peer subsystems and waits for the response
messages. Hence, sophisticated features of P2P
networks are not required.
Basic operations in this network structure are as
follows:
Join to Network: The information of the new
DSWSD-S is introduced to one of the peer DSWSD-
S. Peer DSWSD-S sends the specifications of other
DSWSD-Ss to the new DSWSD-S. New DSWSD-S
sends an introduction message that contains its
details to each of other DSWSD-Ss. Every peer is
expected to send an acknowledge message to the
new DSWSD-S.
Leave from Network: Any peer can leave the
network in two different ways: explicitly by
initiating a resign procedure or implicitly, or without
informing any other peers. In resign procedure
resigning DSWSD-S sends resign message to other
peer DSWSD-Ss then immediately leave the
network without waiting any acknowledge message.
In the former case, if a DSWSD-S peer detects a
down peer with down time above a previously set
threshold, then it sends other peers down message
about this peer. DSWSD-S peer that receiving down
message updates peer routing table.
Related Node Routing: When a service receives a
discovery request which is not related its own
ontology, it forwards the request to the peers on the
basis of the information in peer routing table and
begins waiting. If there were any discovery results,
results are redirected to the requester peer
DSWSD-S.
3 SERVICE ACQUISITION
AND VALIDATION
The basic step in service acquisition is collecting the
addresses of Web services. This step is similar to job
performed by search engines. They traverse the
whole web and collect the necessary data. Since, in
SERVICE ACQUISITION AND VALIDATION IN A DISTRIBUTED SERVICE DISCOVERY SYSTEM CONSISTING
OF DOMAIN- SPECIFIC SUB-SYSTEMS
95

this work, the objective is not building a crawler, but
rather effective service discovery, initially, we
decided to use an off-the-shelf crawler and examined
several existing open source crawlers, including
Java-based crawlers and .net based crawlers. In
order to provide a seed to the crawlers, we used the
results obtained from search engines (Google and
Yahoo). However, this approach could retrieve only
a few web service description files. For this reason,
we decided to try other ways to collect service
descriptions.
As an alternative approach, we considered using
search engines directly in finding Web service
description sources. In this approach, we used the
filtering capability of search engines of Google and
Yahoo for file extensions while searching Web
content. Most of the Web service descriptions are
constructed in WSDL (wsdl, 2001) and stored with
“.wsdl” extension. At this point, as described in the
next paragraph in more detail, we made another
extension. Since “.wsdl” file type filtering did not
yield satisfactory enough results, we preferred to use
“.asmx” in file filtering.
Web service description file in WSDL format
contains many constructs. Indeed, these constructs
are XML based text files. Therefore, WSDL file
should be parsed in order to retrieve advertised web
services provided on a given URL. To this aim, we
implemented a customized WSDL parser that
extracts inputs, output, documentation of web
service and complex types defined as either input or
output. We have observed that not all of the
descriptions and well-formed and the parser return
error messages for such cases. When we examined
the error messages, we have seen three types of
problems with the service descriptions:
 Service description can be specified in an
earlier version of WSDL.
 Service description can be a malformed
WSDL document.
Service description is not given in WSDL at
all; it is specified as an ordinary html
document with wsdl extension.
The validation task is applied on the parsable
descriptions. The next concern is to identify the
input types to provide suitable values for parameters.
Input types may be in primitive or complex
structure. Handling the primitive typed parameters is
straightforward, by using the following rules:
 For primitive types of enumeration, character,
boolean, integer and floating point number, the
input is set as the value1.
 Check string type parameter names are checked
to see if it contains the word “date”. If it
contains the word “date”, use the date that has
the same month and day value and that is
closest to the current date as the parameter
value.
 For other strings, the parameter is set to be a
simple text such as “text”.
When compared with primitive types, handling the
complex typed parameters is a challenging issue.
While dealing with complex types, we have
encountered the following situations that problems:
 complex type of a parameter may refer to
another complex type
 parameter type is an array of complex type
 parameter type is a reference type.
 parameter type is an interface.
 parameter type is a nullable primitive.
For each of these situations, we used the following
solutions:
 In order to create an instance of a complex
typed parameter, Reflection utility of .net
framework is used. Existence of a reference to a
previously created type may lead to infinite
loop. For this reason, constructed complex type
instances are kept in a list and we refer to this
list for setting the values when necessary.
 For complex typed arrays, we create one
dimensional array having a single place. Its
instance is constructed and set by using the
procedure described above.
 When parameter is of a reference type, only the
instance of top level type is created and we do
not set any values to its attributes.
 In order to provide Interface type parameters,
all the complex types that are previously
defined are searched until. When a type
implementing that interface is found is found,
type instance is created and values are set.
 Since primitive types cannot be set to null due
to its nature, instead of a primitive typed
instance, an instance of object type that
corresponds to the given primitive type is
constructed and its value is set.
After solving all the problems with the complex
types, web services are invoked one by one and the
responds are analyzed to find the validated (active)
services. For the filtering with “.wsdl” files,
although a long source URL list is obtained, most of
the descriptions cause parse errors. Among the
parsable ones, the number of validates services is
ICEIS 2010 - 12th International Conference on Enterprise Information Systems
96

Table 1: wsdl vs asmx in general.
Valid WSDL
Ratio
Valid Service
Ratio
Car – wsdl 60 % 23 %
Car – asmx 35 % 93 %
Aeronautic - wsdl 23 % 48 %
Aeronautic - asmx 43 % 96 %
Table 2: wsdl vs asmx details.
Car – wsdl Car – asmx Aeronautic – wsdl Aeronautic – asmx
Total # of URLs 50 161 111 162
Valid # of URLs 30 58 26 71
Invalid # of ULRs 20 103 85 91
Total # of Service 772 632 82 1852
Valid # of Service 213 588 39 1786
Invalid # of Service 559 44 43 66
Duration 37 Mins. 17 Mins 36 Mins 48 Mins
quite few, showing that these services are not active
and callable any more. Since filtering “.wsdl” files
approach did not produce satisfactory results,
considering that there must be much more Web
service available then validated set and some of
them can also be dynamically created by web
servers from .asmx files with “wsdl” parameter as
query string (---.asmx?wsdl), we decided to use
“.asmx” file filtering in address acquisition phase.
When the service provider address list obtained with
this new filtering was evaluated, we have observed
that the number parse problems decreased and the
number of active services increases. The comparison
of these two different file filtering attempts is
presented in Section 4, Table 2.
4 EXPERIMENTAL RESULTS ON
SERVICE ACQUISTION AND
VALIDATION
In this section we present results of the service
acquisition and validation on real web services. In
these experiments, we used “car” and “aeronautic”
domains for the domain-specific service crawling as
described in Section 2.1.1. We used the results of
single runs and each run is performed on the same
computer with 2.73 GHz Core2Duo processor under
the operating system Windows 7.
As described in Section 3, using different file
filtering gives different performance results. In
Table 1, comparison of using “.wsdl” and “.asmx”
file filtering is presented. As seen in the results, the
percentage of addresses obtained with “.wsdl”
filtering changes with respect to the ontology.
However, from the aspect of active web services,
there is a consistent result showing that there is a
higher number of active web services acquired with
“.asmx” filtering.
More detailed results of this experiment are
presented in Table 2. It is observed that for both
ontology, “.asmx” file filtering acquires more
service provider addresses than “.wsdl file filtering”
The number of services under “.asmx” files is more
than “.wsdl” files both in percentage and magnitude.
As an expected result, it is observed that the runs
take longer time for “wsdl” files due to existence of
high number offline web services. DSWSD-s tries to
connect to a web server and waits until time-out
duration, if the server is offline and this waiting time
increases the overall service acquisition time.
For the active services, we conducted further
examination in order to understand the status of the
services. The results for “wsdl” file filtering for
under car domain are shown in Table 3. As a result
of this examination, active services are further
grouped under Successful Invocation, Suitable
Parameter, Unsuitable Parameter, Unsuccessful
Invocation, Argument Error and Insecure
Invocation.
For the active services, we conducted further
examination in order to understand the status of the
services. The results for “wsdl” file filtering for
under car domain are shown in Table 3. As a result
of this examination, active services are further
grouped under Successful Invocation, Suitable
Parameter, Unsuitable Parameter, Unsuccessful
Invocation, Argument Error and Insecure
Invocation. Successful Invocation indicates that the
SERVICE ACQUISITION AND VALIDATION IN A DISTRIBUTED SERVICE DISCOVERY SYSTEM CONSISTING
OF DOMAIN- SPECIFIC SUB-SYSTEMS
97

Table 3: Run Details.
Number of total URLs 50
Parsable URLs 30
Active Services 213
Successful Invocation 167
Suitable Parameter 34
Unsuitable Parameter 133
Unsuccessful Invocation 46
Argument Error 31
Insecure Invocation 15
Offline Services 559
Unparsable URLs 20
web service was invoked successfully and returned
appropriate outputs. In case of unsuitable input
parameters, “soapException” is obtained. For
example, assume a service with the interface
getProduct(int productId). Possibly, productId is the
key of the record in provider’s database. If the value
assigned to productID is a valid value in the
provider’s database by coincidence, a successful
invocation occurs. Otherwise, soapException is
obtained. Unsuccessful Invocation means that web
service cannot be invoked successfully although
service is online. It occurs when the created
parameter types are not compatible (Argument
Error) or when service requires connection over
https. (Insecure Invocation).
5 RELATED WORK
Web service discovery is a very active research area.
Current studies are generally on Web service
database construction, web service crawler
implementation, semantic web service discovery,
indexing of semantically annotated services, quality
of service (QoS) issues and domain specific web
service discovery.
Al-Masri & Mahmoud (2007a, 2007b, 2008)
works on building a web service crawler. Web
Service Crawler Engine (WSCE) is able to handle
WSDL files and UBRs information concurrently and
store the collected information in a centralized
database called Web Service Storage (WSS).
Gooneratne & Tari (2007, 2008) works on
semantic web service discovery. In their work,
service discovery is done under functional and non-
functional requirements or both.
Kuang, Ying, Wu, Deng and Wu (2007) propose
indexing all ontology-annotated data in registered
services. Based on indexing, composition-oriented
service discovery is proposed.
Xu, Martin, Powley and Zulkernine (2007)
works on QoS attributes of web services and they
propose a QoS based web service discovery
approach. In their work, every web service is scored
according to the feedback of users about the service
performance. The reputation manager collects
feedback, calculates reputation scores, and updates
these scores in the rating database.
Ma, Wang, Li, Xie and Liu (2008) works on QoS
attributes of web services in a semantic point of
view. They propose a semantic QoS-aware
discovery framework that makes use of constraint
programming.
Rocco, Caverlee, Liu and Critchlow (2005)
propose service class driven dynamic data source
discovery on domain specific web service. They
search “Deep Web” which refers to the dynamic
collection of Web documents that are created as a
direct response to some user query. Our work differs
from the abovementioned studies with its feature of
being a collection of domain-specific subsystems.
This improves the service acquisition and validation
cycle time improving the service suggestion. In
addition, the proposed work improves the previous
studies on several points such as using semantic
matching in querying and including matching value
in the index.
6 CONCLUSIONS
In this work, we present a distributed service
discovery system consisting of domain-specific
service discovery subsystems. In this system,
ontology driven domain specific web service
discoverer sub-systems are executed in parallel to
handle the acquisition of the services. By this way,
service discovery process is accelerated and more
accurate results can be obtained in terms of service
validation, verification and QoS values.
In this paper, while the architecture of the system
and main functionalities of modules and relations
between modules are presented, the emphasis is on
service acquisition and validation steps. These steps
are important for the rest of the modules, since they
provide the input to be processed. Our experimental
evaluations on the service acquisition and validation
show interesting results on the current web services.
It has been observed that the search engines are
better tools than using off-the-shelf crawlers. This
may be due to nature of the seed sets used in the
experiments. However, using search engines
facilitates the acquisition task by removing the need
for trying out different seed sets for satisfactory
ICEIS 2010 - 12th International Conference on Enterprise Information Systems
98

service acquisition. Another important observation is
that currently, using .asmx files for dynamic service
specification is a common practice. For this reason,
“.asmx” file filtering on search engines provides
better service acquisition than “.wsdl” file filtering.
Another observation that supports this finding is that
most of the services acquired under “.wsdl” filtering
are not active any more.
REFERENCES
Al-Masri, E & Mahmoud, QH, 2007, Crawling multiple
UDDI business registries, 16th WWW Conf. Pp. 1255-
1256.
Al-Masri, E & Mahmoud, QH, 2007, WSCE: A crawler
engine for large-scale discovery of web services,
ICWS pp. 1104-1111.
Al-Masri, E & Mahmoud, QH, 2008, Investigating web
services on the world wide web., WWW pp.795-804
ebXML, 2005, Registry services specification v3.0,
http://www.oasis-open.org/specs/index.php
Gnutella, viewed Jan 2010 http://rfc-gnutella.
sourceforge.net/.
Gooneratne, N & Tari,Z, 2008, Matching independent
global constraints for composite web services, WWW
pp. 765-774.
Gooneratne, N, Tari, Z & Harland, J, 2007, Verification of
Web Service Descriptions using Graph-based
Traversal Algorithms. In Proceedings of the ACM
Symposium on Applied Computing, pages 1385–1392.
Kuang, L, Li, Y, Wu, J, Deng, SG & Wu, Z, 2007,
Inverted Indexing for Composition-Oriented Service
Discovery. ICWS pp.257-264.
Ma, Q, Wang, H, Li, Y, Xie, G & Liu, F, 2007, A
Semantic QoS-Aware Discovery Framework for Web
Services. ICWS pp.129-136
METEOR-S: 2006, viewed Jan 2010, http://
lsdis.cs.uga.edu/projects/ meteor-s/
Morris, R, Karger, D, Kaashoek MF & Balakrishnan, H,
2001, Chord: A Scalable Peer-to-Peer Lookup Service
for Internet Applicaons, Ion Stoica
Rocco, D.; Caverlee, J.; Liu, L.; Critchlow, T. 2005,
Domain-specific Web service discovery with service
class descriptions, ICWS. pp.481 - 488
Sahin, OD, Gerede, CE, Agrawal, D, Abbadi AE, Ibarra,
OH & Su, J, 2005 Spider: P2p-based web service
discovery. 3rd International Conference Service-
Oriented Computing (ICSOC 2005), pages 157–169.
Springer
UDDI, 2004 Version 3.0.2, viewed Jan 2010, descriptions
http://uddi.org/pubs/uddi_v3.htm.
WSDL, 2001, version 1.1, viewed Jan 2010, descriptions
Mar 2001, www.w3.org/TR/wsdl.
Xu, Z, Martin, P, Powley, W & Zulkernine, F, 2007,
Reputation-Enhanced QoS-based Web Services
Discovery. ICWS pp.249-256.
SERVICE ACQUISITION AND VALIDATION IN A DISTRIBUTED SERVICE DISCOVERY SYSTEM CONSISTING
OF DOMAIN- SPECIFIC SUB-SYSTEMS
99