Abstract—While Web Services already provide distributed operation execution, the registration and discovery with UDDI is still bad on a centralized design. In this paper we show to build a distributed discovery rvice, bad on a peer-to-peer infrastructure. Furthermore, we u DAML-S rvice descriptions to provide enhanced mantic arch capabilities. Our prototype implementing this approach bad on the peer-
to-peer infrastructure Edutella is described.
Index Terms—Semantic Web Services, peer-to-peer networks, Web Service registry.
I.INTRODUCTION AND MOTIVATION
ODAY distributed computing is connected with the idea of Web Services. The are (parts of) programs that can be accesd over a network using well defined protocols. An important aspect is that the interactions should be done automatically by computers. Currently one of the main problems is the locating of Web Services which provide the desired functionality.
This paper is divided as follows. In ction II we give an overview of UDDI and WSDL. In ction III we describe our idea and an example of a peer-to-peer and mantic web bad registry for Web Service
descriptions. We show concrete parts of our prototype that we implemented bad on our Edutella-infrastructure in ction IV.
II.B ACKGROUND:UDDI AND WSDL
A.Running Example
Throughout this paper we will u as example Web Services for image downloads. MD Smith is a medical instructor. He wants to prepare a lecture about leukemia. For this purpo he needs images of leukocytes (white blood cells). He wants to include a diagram showing how the leukocytes are structured. He is a bit peculiar about the image quality, and therefore insists of having the diagram as vector graphic.
Today there are probably hundreds of medical image Manuscript received February 17, 2003.
All Authors are with Learning Lab Lower Saxony, University of Hanover, Expo Plaza 1, D-30539, Hanover, Germany.
{thaden, siberski, nejdl}@learninglab.de. databas available via internet, many of them owned by universities. Besides, the graphics needed could also be provided by general image providers, e. g.
press agencies or commercial image archives. Currently most of them are accessible via HTML ur interface only, but for our example we assume that access to the image archives via Web Services is already provided.
B.UDDI Design and Infrastructure
To archive Web Service integration, one must be able to locate specified rvices. The typical process is shown in figure 1. In a first step a Web Service provider publishes his offerings (publish). After that clients can arch for a Web provider that offer the asked functionality, so it can connect to and u the Web Service (bind).
Universal Description, Discovery and Integration (UDDI) is a specification for business registries from Ariba, IBM and Microsoft [1]. It is a central (possibly replicated) registry which contains information about business and their provided Web Services (private registries are possible, but lead to drawbacks as discusd in the next ction). Centralized approaches combined with replication have many drawbacks, e. g. poor scalability and less consistency on large registries. UDDI can be en as typical yellow pages, pointing to registered Web Services which can be located elwhere. UDDI is a registry; it does not specify any kind of registry to manage models or even schemas or metadata. Thus, relationship information cannot be provided.
An item in a UDDI registry consists of the following parts:
•businessEntity: Describes a business or other organization that typically provides Web Services.
•businessService: Describes a collection of related
A Semantic Web bad Peer-to-Peer Service
Registry Network
Uwe Thaden, Wolf Siberski, Wolfgang Nejdl
T
Fig. 1. Web Service registration and discovery
Web Services offered by an organization described by a businessEntity.
• bindingTemplate: Describes the technical
information necessary to u a particular Web Service.
• tModel: Describes a “technical model” reprenting
a reusable concept, such as a We
b Service type, a protocol ud by Web Services, or a category system. • publisherAsrtion: Describes, in the view of one businessEntity, the relationship that the
businessEntity has with another businessEntity.
• subscription: Describes a standing request to keep
track of changes to the entities described by the subscription. Only tModels enable a arch in the UDDI-registry that comes near to our idea (however, one will barely arch for a Web Service bad on the providing company).
For putting in a new Web Service one has to know the exisiting tModels. Furthermore any kind of relation between tModels cannot be expresd; a new tModel can’t be described as a new version of an existing one. Thus, any hierarchy is impossible.
Furthermore, the arch facility is limited; one can arch by keywords, but cannot ask for “something similar” since UDDI does not provide a vocabulary (ontology). The tModels
as well as the WSDL descriptions are not stored in the UDDI registry, only the names and some describing keywords. Following the UDDI-specification it is provided to 1. Find Web Services implementations that are bad on a common abstract interface definition.
2. Find Web Services providers that are classified
according to a known classification scheme or identifier system.
A tModel reprents an abstract rvice type (a generic reprentation of a registered rvice) in the UDDI registry. Each Web Service is categorized according to a defined list of rvice types.
Each tModel has a name, an explanatory description and an unique identifier. The tModel name identifies the rvice, e. g. “Online order placement”. The description gives additional information, e. g. “Place an order online”.
In our example the ur wants to find a Web Service that allows downloading images in vector format. Using a UDDI-tModel only allows describing the first part of that, but it is not possible to describe that a Web Service should have a vector graphic output.
C. WSDL Service Description The rvices registered with UDDI are described in the
Web Service Description Language (WSDL, [2]). This language allows describing all technical information needed
to access the rvice, including its operations with parameters and results. For example, an image download rvice could be described as follows:
<definitions name="MyImageArchive"
targetNamespace="example/wsdl/MyImageArchive">
…
<message name="ImageArchive_arch"> <part name="keyword" type="xsd:string"/>
<part name="imageType" type="xsd:string"/> </message> <message name="ImageArchive_archRespon">
<part name="image" type="xsd: ba64Binary" "/>
</message> </message> […]
<portType name="ImageArchive">
<operation name="arch" parameterOrder="keyword imageType">
<input message="tns:ImageArchive_arch"/>
<output message="tns:ImageArchive_archRespon"/> </operation> </portType>
[…] <rvice name="MyImageArchive"> <port name="ImageArchivePort"
binding="tns:ImageArchiveBinding">
<soap:address
location="example/imagervice:9765"/> </port> </rvice> </definitions>
WSDL perfectly describes how a rvice is called, and thus (in combination with SOAP) facilitates platform and
programming language independent remote operation execution. However, WSDL is less suitable to describe what a rvice does. For example, while we can express that the rvice returns binary dat
a in ba64 format, we can’t state that this data is in fact the reprentation of an image.
Registry Company Y Image Appointment ...
Im age Image Service
Image Image
Text Video Image Cour Enrollment Registry
Fig. 2. Services registered at central registries
III.P EER-TO-P EER AND DAML-S
A.Connecting Registries with Peer-to-Peer Technology While the idea of UDDI is a centralized storage of Web Service descriptions, currently most providers of Web Services tend to t up their own so called private registries. Several commercial UDDI registry products are available which help business to do that. A new trend is to integrate UDDI registry features in more general company directories (e. g. Microsoft Active Directory, Novell eDirectory [3]). This will further enforce the trend to company-wide UDDI registries and limit the usage of central public registries.
This situation makes it difficult to discover rvices. Technically, it would be possible to replicate registry information from all private registries to the public central nodes. However, this needs a replication contract between both registry providers, and manual system administration for each new private registry. Therefore, while technically possible, practically replication from private to public UDDI registries doesn’t occur.
To alleviate this problem, we propo moving from a central design to an (de facto already existing) distributed approach by connecting private registries with peer-to-peer technology. The power of the internet comes topmost from moving from centralized solutions to non-structured information bad at each computer connected to the net. Using peer-to-peer architecture follows this idea and adds value by creating a virtual global registry from all connected local registries [4].
Peer-to-peer enables the Web Service registries. Thus, companies as well as universities can build their own Web Service registries which are maintained by themlves. Being a peer in a P2P-network makes it easy to arch all local registries.
The idea of a P2P network of UDDI rvice registries has already been mentioned in [5]. [6] have propod to combine Web Services, peer-to-peer and Semantic Web technology to build the next-ge
neration rvice driven systems. An ontology-bad peer-to-peer topology suitable for rvice discovery has been described in [7].
Speed-R is a project at University of Georgia that developed a distributed registry bad on the current UDDI without mantically enriched Web Service description [8]. While we envision a registry partition where institutions maintain their private registries, the Speed-R system assumes a partition bad on business domains. The most important difference is that we u DAML-S as rvice description language instead of UDDI tModels. This is the prerequisite for enhanced mantic arch capabilities, as described in the following ction.
B.Service Description and Discovery with DAML-S
To support discovery requests beyond a yellow pages-lookup style, mantic matching capabilities are a prerequisite. For our purpo we u DAML-S [9] to describe the provided functionality of a Web Service. A DAML-S description always consists of four parts as shown in figure below.
A resource (here a program (-part)) offers a rvice, thus our Web Service. This rvice is described by three parts: •Service Profile(s)
The specification of what a rvice requires and
provides for an external (…black-box“) view of the
rvice, e. g. required inputs.
•Service Model(s)
The specification of how a rvice works for an
internal (…glass-box“) view of the rvice, e. g. steps
for a transaction.
•Service Grounding(s)
Implementation specific details (e. g. message
formatting).
[10] has shown that DAML-S can be en as an extension of WSDL: It is possible to map a WSDL description to DAML-S, and provide information not expressible in pure WSDL as add-on. In our prototype we therefore u DAML-S to describe rvices. However the standardized Web Service
R egistry Press Agency D
comm
Hospital D
Image
Download
Service
Appointment
<
Im age
Download
Service
Image
Order
Service
Image
Image
substringDownload
Service合肥翻译
Text
Download
Service
Video
Download
英国 公投
Service
Image nload
Cour
Dow nload
Service
Enrollment
Service
P2P Registry Network
Registry
R egistry
Registry
Registry
Fig. 4. DAML-S rvice description overview
architecture is ud to finally interact with a rvice.
In our example we u an ontology describing different types of Web Services. We have a super class defining a rvice at all. Subclass are “ImageService”, subdivided into “DeliverByMailImageService” and “Downloadable-ImageService”. For example, the latter one describes all Web Services that can be ud somehow to receive URLs of images to download (satisfying given criteria like output URL must point to a vector image).
[…]
<rdfs:Class
rdf:about="&rviceType;DownloadableImageService"> <rdfs:subClassOf
rdf:resource="&rviceType;ImageService"/>
<rdfs:label>Downloadable Image Service</rdfs:label>
</rdfs:Class>
[…]
For the Web Service description it’s enough to write down the rvice profile since it holds all necessary information like Web Service description, inputs, and outputs. The interesting parts are shown below:
[…]
<rviceType:DownloadableImageService
rdf:about="&example;MyImageArchive">
artist<profile:input>
<profile:ParameterDescription
rdf:about="&example;Keyword">
<profile:parameterName>
keyword
</profile:parameterName>
[...]
</profile:ParameterDescription>
</profile:input>
<profile:input>
<profile:ParameterDescription
rdf:about="&example;ImageType">
<profile:parameterName> image type
</profile:parameterName>
[...]
</profile:ParameterDescription>
</profile:input>
<profile:output>
<profile:ParameterDescription
rdf:about="&example;Image">
<profile:parameterName>
ImageFound
</profile:parameterName>
<profile:restrictedTo rdf:resource="ℑVectorImage" /> </profile:ParameterDescription>
</profile:output>
</rviceType:DownloadableImageService>
[…]
Bad on this kind of description, it is possible to arch for all rvices of type DownloadableImageService which have an output of type VectorImage. Such a query wouldn’t have been possible with tModels only, becau it isn’t possible to characterize parameter and result types (except assigning XML Schema types).
An effective matching algorithm for DAML-S between rvice descriptions and requests has been developed [16]. C.Implementation
For our implementation we reu the existing project Edutella. Edutella is a general peer-to-peer infrastructure for storing, querying and exchanging metadata [11]. It is built on the open source project JXTA, a framework which provides basic peer-to-peer network features [12]. Edutella can connect highly heterogeneous peers (heterogeneous in their uptime, performance, storage size, functionality, number of urs etc.). To achieve the desired interoperability, it is crucial to adhere to standards [13]. Therefore Edutella is bad on metadata standards defined by the SemanticWeb initiative of the WWW Consortium [14], namely RDF and RDFS. Each Edutella peer can make its metadata information available as a t of RDF statements, suitable for describing distributed resources.
This information can be queried using the Edutella query exchange language (QEL). QEL defines veral capability levels, starting with simple conjunctive queries (which allow a query-by-example style of request) up to capabilities comparable to query languages of state-of-the-art relational databas. QEL queries are distributed to all relevant peers (e [15] for the query distribution mechanism), and matches nt back to the query originator. We have implemented wrappers to veral RDF stores which translate the QEL queries to the backend-specific query language, and can therefore integrate various kinds of backends into the network, from files to RDBMS systems and inferencing databas.
As DAML-S descriptions are already provided in RDF format, Edutella is perfectly suited to provide them. To arch
Fig. 5. Image ontology
for rvices, a peer nds an appropriate QEL query to the network.
point是什么意思For simple queries, QEL provides a query-by-example syntax, where arbitrary nodes in the RDF graph are tagged as variables. The discovery query for all rvices of type DownloadableImageService which have an output of type VectorImage looks as follows:
<edu:QEL1Query rdf:about="#rviceQuery"> <edu:hasVariable rdf:resource="#Service"/> <edu:hasVariable rdf:resource="#Output"/>
<edu:hasResultType rdf:resource="&edu;TupleResult"/> </edu:QEL1Query>
<edu:Variable rdf:about="#Service"/> <edu:Variable rdf:about="#Output"/>
<rviceType:DownloadService rdf:about="#Service">
<rdf:type rdf:resource="&rviceType;DownloadService"/> <profile:output>
<profile:ParameterDescription rdf:about="#Output">
<profile:restrictedTo rdf:resource="ℑVectorImage" /> </profile:ParameterDescription> </profile:output>
</rviceType:DownloadService>
As described, such a query can be procesd by already existing peer implementations. For a complete implementation we only had to add a registration facility for
new or changed rvices. Services have to provide their DAML-S description when registering. This information is stored as RDF and can be accesd using the default query
sm是什么mechanism. The WSDL description is stored as well; rvice clients can u it to access discovered rvices via SOAP.
Our prototype ur interface for discovery allows arching
for rvices by lecting concepts from veral ontologies. For rvices found a simple ur interface is generated, using the
WSDL information, and the rvice can be called manually. IV. C ONCLUSION AND F URTHER W ORK
We have shown that introducing peer-to-peer technology for Web Service registries provides the advantages of easy tup and maintenance of local registries while prerving the ability for global rvice discovery. This is achieved by distributing discovery requests to all registry peers within the
network. By introducing DAML-S rvice profile descriptions in combination with ontologies we gain enhanced mantic
query capabilities.
It would be interesting to combine our discovery mechanism with the work done by Hendler et al. to create composite Web Services [17]. Service elements could be found using the registry network and then plugged-in into the designed composite rvice.
Currently only discovery is done via the peer-to-peer network. The next logical step would be to let the network
also handle the rvice binding and execute the rvices on behalf of the query originator. Thus, in our example, a client would nd the rvice parameters together with the rvice discovery request, and receive the images directly instead of rvice descriptions.
A CKNOWLEDGEMENT
Stefan Decker propod the initial idea for distributed Web Services in one of our discussions, which finally led to the work described in this paper.
R EFERENCES
[1] “The UDDI Specification V3”, www.uddi
[2] Roberto Chinnici, Martin Gudgin, Jean-Jacques Moreau, Sanjiva
Weerawarana, Web Services Description Language (WSDL) Version 1.2, www.w3/TR/wsdl12/
[3] Cathleen Moore. “Novell rolls out UDDI rver”. Infoworld article.
/article/02/12/11/021211hnnoveluddi_1.html [4] Andy Oram (Ed.), “Peer-to-Peer: Harnessing the Power of Disruptive
Technologies”, O’Reilly, 2001
[5] Graham Glass. Web Services . Prentice Hall, 2002
[6] A. Maedche, S. Staab, “Services on the move – Towards P2P-Enabled怎么样克服紧张
Semantic Web Services”, Tenth International Conference on Information Technology and Travel & Tourism, ENTER 2003, Helsinki, 2003.
[7] Mario Schlosr, Michael Sintek, Stefan Decker, Wolfgang Nejdl,
“HyperCuP - Hypercubes, Ontologies and Efficient Search on P2P Networks”, International Workshop on Agents and Peer-to-Peer Coputing,
Bologna, Italy, July 2002
[8] Kaarthik Sivashanmugam, Kunal Verma, Ranjit Mulye, Zhenyu Zhong.
“Speed-R: Semantic P2P Environment for diver Web Services
Registries”. webster.cs.uga.edu/~mulye/SemEnt/final.html [9] DAML-S Coalition, “DAML-S: Web Service Description for the Semantic
Web”, First International Semantic Web Conference, Sardinia, Italy, 2002. [10] M. Paolucci, T. Kawamura, T. R. Payne, K. Sycara. “Importing the
Semantic Web in UDDI”. Workshop on Web Services, e-Business, and the Semantic Web (WES). Fourteenth International Conference on Advanced
Information Systems Engineering, Toronto, Canada, 2002.
[11] W. Nejdl, B. Wolf, C. Qu, S. Decker, M. Sintek, A. Naeve, M.Nilsson, M. Palmér_ and T. Risch, “Edutella : A P2P Networking Infrastructure Bad on RDF”. Eleventh International World Wide Web Conference
(WWW2002). Available at: edutella.jxta/reports/edutella-whitepaper.pdf. [12] L. Gong, “Project JXTA: A Technology Overview”, Sun Microsystems,
2001. Available at www.jxta/project/www/docs/
TechOverview.pdf. [13] R. Dornfest, and D. Brickley, “The Power of Metadata”, in Peer-to-Peer: Harnessing the Power of Disruptive Technologies
[4]. Available at /pub/a/p2p/2001/01/18/metadata.html
[14] T. Berners-Lee, J. Hendler, O. Lassila, “The Semantic Web”, Scientific
subtletyAmerican, May 2001. Available at
/2001/0501issue/0501berners-lee.html.
[15] Nejdl, Wolfgang; Wolpers, Martin; Siberski, Wolf; Schmitz, Christoph;
Schlosr, Mario; Brunkhorst, Ingo; Lör, Alexander. “Super-Peer-Bad Routing and Clustering Strategies for RDF-Bad Peer-To-Peer
Networks”. Accepted for Publication at 12th International World Wide Web
Conference (WWW2003), Budapest, Hungary, May 2003. [16] M. Paolucci, T. Kawamura, T. R. Payne, K. Sycara. “Semantic Matchingcal
of Web Services Capabilities”. First International Semantic Web
Conference, Sardinia, Italy, 2002.
[17] E. Sirin, J. Hendler, B. Parsia, …Semi-automatic Composition of Web
Services using Semantic Descriptions“. .Accepted to Web Services:
Modeling, Architecture and Infrastructure workshop in conjunction with ICEIS2003.
