Computing and Informatics, Vol. 27, 2008, 93–118
UNDERSTANDING SEMANTIC AWARE GRID
MIDDLEWARE FOR E-SCIENCE
Pinar Alper, Carole Goble
School of Computer Science, University of Manchester, Manchester, UK
e-mail: {penpecip, carole}@cs.man.ac.uk
Oscar Corcho
School of Computer Science, University of Manchester, Manchester, UK
&
Facultad de Informa´tica, Universidad Polite´cnica de Madrid
Boadilla del Monte, ES
e-mail: ocorcho@cs.man.ac.uk
Revised manuscript received 11 January 2007
Abstract. In this paper we analyze several semantic-aware Grid middleware ser-
vices used in e-Science applications. We describe them according to a common
analysis framework, so as to find their commonalities and their distinguishing fea-
tures. As a result of this analysis we categorize these services into three groups:
information services, data access services and decision support services. We make
comparisons and provide additional conclusions that are useful to understand bet-
ter how these services have been developed and deployed, and how similar ser-
vices would be developed in the future, mainly in the context of e-Science applica-
tions.
Keywords: Semantic grid, middleware, e-science
1 INTRODUCTION
The Science 2020 report [40] stresses the importance of understanding and manag-
ing the semantics of data used in scientific applications as one of the key enablers of

94 P. Alper, C. Goble, O. Corcho
future e-Science. This involves aspects like understanding metadata, ensuring data
quality and accuracy, dealing with data provenance (where and how it was pro-
duced), etc. The report also stresses the fact that metadata is not simply for human
consumption, but primarily used by tools that perform data integration and exploit
web services and workflows that transform the data, compute new derived data, etc.
This vision of e-Science is also presented in [41], where the authors foresee a high
degree of seamless automation, with flexible collaborations and computations on
a global scale. Here the Semantic Grid is defined as an extension of the current Grid
where information and services are given well-defined meaning, better enabling com-
puters and people to work in cooperation, and is considered as a suitable approach
for the new challenges. In such an environment, the use of explicit knowledge and
metadata underpins sharing and brings interoperation, flexibility and extensibility
to Grid systems.
System development in the Semantic Grid has evolved from a set of pioneer
applications, with ad-hoc developments, to a phase of systematic investigation, with
more logical, consistent and ordered approaches and where know-how transfer to
middleware and applications has started. A non-exhaustive list of examples of sys-
tems developed in the first exploratory phase: Geodise [34], myGrid [35], Comb-
e-Chem [24], and CoAKTinG [38]. All of them use semantic annotations of re-
sources to fulfill their requirements (exposing metadata, reasoning with knowledge
and metadata, etc.). Besides, they all use semantic technologies, but these tech-
nologies are not especially adapted to be used in Grid middleware platforms and
applications.
Another non-exhaustive list of systems, most of them under development, in
the second phase: caGrid [15], BIRN [16], the Semantic Reef [37], the OntoGrid’s
QUARC system [36], and GEON [39]. These systems have been developed fol-
lowing a more systematic approach, are more tightly integrated with existing Grid
infrastructure and in some cases they have been developed according to any of the
Semantic Grid architectures recently proposed, such as S-OGSA [19], InteliGrid’s
architecture [42] and NextGrid’s architecture [43].
In the second phase several Grid middleware systems have been also partially
or completely re-implemented with semantic technologies, or have been wrapped so
that they provide semantic-aware interfaces. This paper is specifically focused on
these services, which we call Semantic-Aware Grid Middleware Services (SAGMS).
In the following section they will be analysed and compared, delving into their role
in Grid middleware, their interplay with services that give them semantic support,
and the means with which they have been developed.
The advantage of focusing only on middleware services is that we do not have to
make assumptions about the usually complex domains of complete Semantic Grid
applications, what could hinder the descriptions and results presented in the paper.
Middleware systems are easily understood and have clearly-defined specifications
and functionalities. This study can be then extrapolated to complex applications.
In summary, the purpose of this paper is to provide insight onto the features,
commonalities and differences of different Semantic-Aware Grid middleware services,

Understanding Semantic Aware Grid Middleware for e-Science 95
including the design decisions behind them. We try to use these experiences to guide
future system designs and developments.
The paper is structured as follows: Section 2 gives some background about
Grid architectures and specifications, which will be useful to understand the sys-
tem descriptions provided later. Section 3 describes the analysis framework used
to describe and compare the Semantic-Aware Grid Middleware Services (SAGMS)
selected. This framework is not aimed at evaluating the decisions made, but for
understanding such decisions. Section 4 describes each of the systems that are part
of our study. Section 5 summarises the results of applying the framework to these
services, categorises them according to their functionalities, and provides compari-
sons between them. Finally, section 6 summarises the work done and the results
obtained.
2 BACKGROUND: GRID SYSTEMS AND THE OPEN GRID
SERVICE ARCHITECTURE
In the end of 2000, the Grid community formed the Global Grid Forum (now Open
Grid Forum – OGF) to lead their standardisation efforts. The convergence between
Web Services and Grid computing led to the proposal of the Open Grid Service
Architecture specification (OGSA) [12]. OGSA defines the Grid architecture by
outlining the requirements of the Grid (based on the needs of relevant use cases),
and identifying the major groups of service capabilities to deliver these functionali-
ties.
OGSA identifes six service groups: data (concerned with the management and
transfer of data resources), resource management (concerned with the management
of physical and logical resources and services, and of Grid infrastructure), execu-
tion management (concerned with the instantiation, management and completion
of units of work), security (concerned with the enforcement of security-related poli-
cies within a virtual organization), self-management (concerned with resource self-
configuration, self-healing and self-optimization) and information (which acts as
databases of attribute metadata about resources).
Some cross-cutting OGSA requirements are: interoperability and support
for dynamic and heterogeneous environments, resource sharing across orga-
nizations, optimization, Quality of Service (QoS) assurance, job execution, data
services, security, administrative cost reduction, scalability, availability, and ease of
use.
OGF actively works on key infrastructure services and protocols that make up
a foundation for higher level OGSA capabilities. These foundational standards
are related to naming, security, state, notification, transactions and orchestrations.
Specifications can be found for all of them, among which we can highlight those re-
lated to state. Two alternatives exist for supporting stateful resources: WSRF [13]
(Web Services Resource Framework), which is composed of four specifications: WS-
ResourceProperties, WS-ResourceLifeTime, WS-ServiceGroup, and WS-BaseFaults,

96 P. Alper, C. Goble, O. Corcho
and is being standardised by OASIS; and WS-Management, composed of WS-
Transfer, WS-Eventing, WS-Enumeration, and WS-ManagementCatalog, and pro-
posed by a group of companies, including Microsoft, Intel, Dell, AMD and Sun Mi-
crosystems. The most relevant OGSA-observant middleware platforms are: gLite [2],
UNICORE [3], Globus [1] and OMII [4]. They all provide a set of middleware func-
tionalities, in the form of Web or non-Web service interfaces, which can be classified
in any of the aforementioned OGSA service groups.
3 AN ANALYSIS FRAMEWORK FOR SEMANTIC-AWARE GRID
MIDDLEWARE SERVICES
In this section we describe our framework for the analysis of Grid middleware ser-
vices implemented or extended with semantic technologies. As aforementioned, this
analysis framework does not aim at ranking these implementations, but at provid-
ing insight on the design and implementation decisions taken to build them, so that
lessons learned can be derived and used in similar future developments.
The analysis is based on a two-dimensional set of features, shown in Table 1.
The first dimension focuses on aspects related to Grid Middleware, Knowledge and
Metadata. For each feature group, we focus on its most relevant aspects (WHAT)
and on how these are implemented (HOW), as follows:
• In the Grid Middleware dimension we analyse the service purpose, including its
(added-value) functionality, the standards used and the status of the software,
how the service was (re-)developed, the standards used or extended, etc.
• In the Knowledge dimension, we focus on the semantic capabilities used, on
their lifecycle (from its creation to its use and disposal), and on their role and
coverage in the system. Besides, we analyse how knowledge entities are accessed
and used, and which representation languages are used to express them.
• Similarly, in the Metadata dimension we analyse the metadata capabilities used,
their lifecycle and their role and coverage, and we analyse how metadata is
accessed and used, and the representation languages used to express it.
WHAT HOW
Grid Middleware Purpose System Development
Knowledge Knowledge Entities Knowledge Technologies
Metadata Metadata Entities Metadata Technologies
Table 1. Dimensions for the analysis of Semantic-Aware Grid Middleware Services
For each dimension, the framework proposes the set of questions or discussion
topics that are identified in the following sections.

Understanding Semantic Aware Grid Middleware for e-Science 97
3.1 Grid Middleware Features of SAGMs
Purpose of the system. In this category we identify the functionalities of the
system from a Grid perspective. The items in this category are:
1. OGSA Service Group, namely Data, Security, Information, Execution,
Resource Management, and Optimization.
2. Functionality. The exact functionality that the system performs in the
context of the aforementioned service group.
3. Added-Value Functionality. Does the service provide new functionality
that did not exist before or is it providing better or flexible ways to deliver
a previously existing functionality?
4. Implementation Standards. The standards that the system implements
(e.g. OGF models and protocols).
5. Software Status. The current status of the software (production, beta,
alpha, etc.).
System Development. Here we identify how the service has been developed.
1. Approach to Obtain Semantic Awareness. Has the system been de-
veloped from scratch or by re-working or extending an an existing Grid
component to make it semantically-aware?.
2. Re-Working method. In the case of re-working there are several ap-
proaches: a. Non-intrusive extension (e.g. client side wrapping). b. Mini-
mally intrusive extension (e.g. interface extensions that might require a re-
compilation without changing the code of the original Grid component).
c. Intrusive extension (e.g. extending protocols or changing code).
3. Extensions to Standards. Has any standard been extended during the re-
factoring? These extensions are normally done using the extensibility points
that most Grid middleware standards provide (e.g. xsd:any fields in most
OGF specs), which in the case of SAGMS could be used to represent or
carry semantic add-ons.
4. Service Orientation. Is a service-based approach adopted to deliver the
service functionality and access metadata and knowledge?
3.2 Knowledge Model Features of SAGMs
Knowledge Entities. We capture the knowledge requirements of the service.
1. Required Knowledge Models. Systems can operate over a pre-specified
knowledge model (i.e. a specific ontology), be agnostic to the knowledge
models (the system is generic and can use any), or not use any form of
knowledge model.
2. Knowledge Model Lifecycle. If it exists, the change rate of the knowledge
model and the type of changes (e.g. mostly additions, rarely updates etc.).

98 P. Alper, C. Goble, O. Corcho
3. Knowledge Model Coverage. The coverage, focusing on whether they
are models of a domain or they are used to represent dynamic aspects such
as conditions and control of the system (e.g. business rules).
4. Knowledge Model Generation. Whether the service supports the know-
ledge provisioning process or it is agnostic to it (that is, it assumes that they
have been created, no matter by whom). In the first case, we focus on the
nature of the generation process: automatic, semi-automatic, manual, etc.
5. Dependency levels to knowledge. A service can have different depen-
dency levels to knowledge, such as mission critical or optional.
6. Security. Whether any security requirements can be applied for accessing
and using knowledge.
Knowledge Technologies. In this category we identify the knowledge technolo-
gies used in the service design and deployment.
1. Knowledge Representation Languages. Which languages and associa-
ted inference techniques have been used within the service implementation.
Examples are RDFS [31], OWL-DL [30], SWRL [27] etc.
2. Knowledge Management Tools and Infrastructures. The knowledge
management tools used by the service. Examples are reasoners and rule en-
gines (e.g., FaCT [21], Jess [22]), RDF libraries (e.g., Sesame [46]), Jena [8]),
OWL2Java source code generators (e.g., Jastor [23]), etc.
3. Knowledge Model Access. The means to access and use knowledge mo-
dels, which can be local or remote, and use Grid-compliant tools or not. For
example, an ontology available in a URL can be accessed and used remotely
with specialized ontology services or downloaded and used locally.
3.3 Metadata Features of SAGMs
The features in this dimension are very similar to those in the knowledge dimension:
Metadata Entities. We capture the metadata requirements of the service.
1. Metadata based on a Knowledge Model. Metadata can be based
on a knowledge model or not.
2. Metadata Lifecycle. The types of changes that metadata suffers (frequent
or non-frequent, additions, updates or removals, etc.).
3. Metadata Coverage. How much system information metadata represents
(current state of the resources used in the application, data for condition
checking, for application business logic, etc.).
4. Metadata Generation. Whether metadata is generated by the system or
externally, and what is the generation process: automatic, semi-automatic,
manual, etc.
5. Dependency levels to metadata. A service can have different dependency
levels to metadata, such as mission critical or optional.

Understanding Semantic Aware Grid Middleware for e-Science 99
6. Security. Whether any security requirements can be applied for accessing
and using metadata.
Metadata Technologies. We identify the metadata technologies used in the ser-
vice design and deployment.
1. Metadata Representation Languages. The languages used within the
service implementation for the representation of metadata. Examples are
RDF [32], XML [45], etc.
2. Metadata Management Tools and Infrastructures. The metadata
management tools used by the service. Typical examples are RDF libraries
(e.g., Sesame, Jena).
3. Metadata Access. The means to access and use metadata (local or re-
mote), and whether Grid-compliant tools are used or not.
4 DESCRIPTION OF ANALYSED SEMANTIC-AWARE GRID
MIDDLEWARE SERVICES
The list of SAGMS selected for our analysis contains some systems developed ex-
ternally to our group, such as S-SRB [44], GRIMOIRES [5], CaBIG data access
services [15] and S-MDS [29], and systems developed by our group, such as Semantic-
OGSA-DAI [33], Grid Meta-Scheduling Service [7] and an ontology-based Role-
Based Authorisation Component [18].
Other Grid middleware services use explicit knowledge and metadata to provide
their functionalities, or their interfaces have been enhanced to use explicit know-
ledge and metadata. Our selection is based on the usage of some of these systems
(GRIMOIRES and CaBIG data access services are used in production Grids) and
in the representativeness of the others (we cover the range of OGSA service groups
where knowledge and metadata has been applied).
All descriptions follow the same structure. First, we provide an overview of the
system, describing what the system consists of, where the system is being applied,
etc. Then we describe the role of metadata in the system, and how it is dealt with.
Then we talk about the knowledge models that these systems use for describing
metadata. Finally, we provide an architectural description of the system once that
it has been enhanced with semantics.
4.1 Semantic Search Engine for the Storage Resource Broker (S-SRB)
The Storage Resource Broker (SRB) is a widely-used system for handling shared
data collections distributed across multiple organizations and heterogeneous storage
systems. As part of the SRB toolkit, the Metadata Catalog system (MCAT) enables
data discovery by providing support for storing, indexing and querying metadata
about the data items stored in SRB.

100 P. Alper, C. Goble, O. Corcho
MCAT is based on a relational model and supports textual and attribute-value
pair based metadata, which is normally enough for most applications. However, in
some cases there is a need for improved search mechanisms that take into account
more information about the application domain or allow more complex queries that
relate different pieces of information. This is the purpose of S-SRB [44].
In S-SRB metadata is represented using the RDF language [32]. S-SRB does
not impose a specific metadata knowledge model, so that any domain knowledge
model can be used, depending on the application that will use it. These domain
knowledge models have to be expressed in the OWL ontology language [30].
S-SRB has been built in as a non-intrusive extension of the existing SRB tooling,
as shown in Figure 1. The system is composed of a semantic server backend and a
web-based client interface. The backend provides access to knowledge models and
metadata, and to their associated reasoning functions, using the Jena API [8] and
the Pellet reasoner [25]. It can be populated via programmatic ways or using the
client interface (an extension of mySRB), which allows uploading domain knowledge
models and transforming MCAT conventional metadata into semantic metadata.
The client interface also allows querying metadata. Users build data search
requests with the predicates and restrictions defined in the corresponding knowledge
models. Those requests are executed by the semantic server backend, which uses
instance reasoning (a.k.a. A-Box reasoning) provided by the Pellet reasoner. Then
the results are returned to the client.
S-
SR
B Ontology
Metadata
Conventional
Metadata
Store (MCAT)
Conventional
UI Client
Extended
UI Client
Annotation
UI Client
Reasoning
semantic search
submit ontology &
metadata
search
Domain Onto
Domain specific
metadata
SRB
file
DB
access
SRB user
Annotation System
submit ontology &
metadata
Fig. 1. S-SRB architecture
4.2 GRIMOIRES Service Registry
GRIMOIRES [5] stands for Grid RegIstry with Metadata Oriented Interface: Ro-
bustness, Efficiency, Security. It is a metadata enabled UDDIv2-compliant [28] web

Understanding Semantic Aware Grid Middleware for e-Science 101
service registry, distributed as part of the OMII Grid middleware suite. GRIMORES
extends the UDDI information model and interface to support the WSDL model for
describing service interfaces.
In GRIMOIRES metadata can be attached to entities in the UDDI and WSDL
models, in the form of key-value pairs or sets of RDF triples. If no metadata is
attached, then it acts as a plain UDDI service registry. The metadata lifecycle can
be managed with WSRF-compliant interfaces, and security is taken into account in
terms of user authentication and authorization at the operation level (e.g. User A
cannot submit a description, User B cannot delete, etc.).
GRIMOIRES does not impose a specific knowledge model to structure meta-
data. It only imposes the basic domain-independent UDDI and WSDL models over
service descriptions, which can be extended with domain-specific parts of service
information. The mechanism to populate and query the GRIMOIRES registry is
through controlled API calls, as shown in Figure 2, although support for freeform
RDQL queries [47] is also planned for the future.
The GRIMOIRES backend is implemented entirely in RDF, in contrast to other
hybrid models where UDDI registries are supported by relational backends and
enhancements are attached as architectural add-ons [10, 9]. The choice for a native
implementation is to avoid the additional communication costs and architectural
complexity of hybrid systems. GRIMOIRES uses Sesame and Jena to manage RDF.
G
RI
M
O
IR
ES
M
et
ad
at
a
Conventional Client
metadata
aware client
semantic search
search
Co
n
ve
n
tio
n
a
l
UD
DI
In
te
rfa
ce
M
et
ad
at
a
In
te
rfa
ce
Annotation System
submit conventional +
enhanced metadata
Domain specific
metadata
R
efe
rs
to

Web service
Web service
Fig. 2. GRIMOIRES architecture
4.3 Semantic Monitoring and Discovery Service (S-MDS)
The Globus Toolkit Monitoring and Discovery System (MDS) is an information
service that consists of two WSRF services for discovering and monitoring resources.
The Index Service provides query/subscription interfaces to resource data, and the
Trigger Service can be configured to take action when specific conditions are met in

102 P. Alper, C. Goble, O. Corcho
the data collected. As in the previous systems, metadata enables improved search
mechanisms that include domain knowledge and use complex queries. S-MDS [29]
extends MDS to create, aggregate and maintain WS-Resource semantic metadata.
The semantic metadata of a WS-Resource is created from its XML-based WS-
Resource properties and transformed into RDF, by a semantic metadata provider
(a. k. a. annotation service), which is associated to each WS-Resource. The meta-
data provider also monitors the WS-Resource, by polling the information from the
property values or by implementing a notification-producer interface, so that the
metadata is always up to date.
Two types of semantic models are used to generate and store semantic metadata
in S-MDS: OWL-S and domain-specific ontologies. The first allows describing ser-
vices associated to resources, while the second is application-dependent and allows
classifying each resource according to its functions.
Domain Specific
Onto
OWL-S &
S-
M
D
S
Metadatasemantic aware
client
Monitor & aggregate
semantic metadata
semantic search
Annotation
ServiceOntology convert
OWL-S &
GLUE
Reasoning
Ontology
Domain specific
metadata
R
efe
rs
to
M
D
S
MDS
Domain Onto
GLUE
WS-Resource WS-Resource
monitor & aggregate
XML metadata
Fig. 3. S-MDS architecture
S-MDS is implemented using the Globus Toolkit 4 aggregator framework. Be-
sides obtaining semantic metadata fromWS-Resource properties, semantic metadata
providers are also in charge of registering that metadata in the semantic metadata
repository, implemented in Jena, which aggregates it and acts as a centralised meta-
data repository of a set of WS-Resources, as shown in Figure 3. S-MDS clients use
RDQL queries to access the semantic metadata. In [29] we can find a description of
a prototype implementation that uses the GLUE ontology for describing computing
resources and that is deployed in a cluster with a local scheduler. This description
shows how all the previous components (manager, semantic metadata provider and
semantic metadata repository can be deployed in the Globus Container.

Understanding Semantic Aware Grid Middleware for e-Science 103
4.4 Semantic Data Integration in the Cancer
Biomedical Informatics Grid (caBIG)
The caBIG project of the U.S. National Cancer Institute (NCI) is a large scale
initiative to deliver IT infrastructure to share distributed data and computing re-
sources in cancer research. caBIG tackles the semantic data integration problem
by the caGrid infrastructure [15] and the caCore toolkit [17]. Here we focus on
the OGSA-DAI compliant caGrid Data Services1, which provide an object-oriented
view of data kept in native formats (mostly relational). Two metadata types are
managed: for describing Data Access service interfaces and for mappings between
a service object model and the native data backend that it refers to.
• The first group of metadata is managed by the cancer Data Standards Repository
(caDSR). caDSR holds annotated OO model fragments (termed Common Data
Elements – CDE) of the data providers, and is maintained by expert curators,
who monitor and supervise model providers according to the caCore’s metadata
provisioning processes. Joining the caGrid as a provider requires devising an
OO model of the data in UML, annotating this model with ontology terms and
storing these annotations as CDEs in the metadata registry. Expert monitoring
of the whole process prevents metadata duplication and increases model re-use.
Besides, caCore provides support for metadata lifecycle management and change
notification. Finally, the caDSR metadata services in caGrid are used by client
applications, not by the data access services themselves.
• The second group of metadata is used by access services during their execution.
CaGRid uses an ad-hoc representation for this metadata.
The object model of a caGrid OGSA-DAI service is annotated with ontologies,
provisioned by an ontology service (the Enterprise Vocabulary Service). Ontologies
are developed using description logic, but deployed only as controlled vocabularies.
As with metadata, ontologies are used by client applications, and not by the caGrid
Data Access services, to query and aggregate data from different data access services.
The system has been implemented by extending OGSA-DAI with a new query
activity that accepts semantic search requests in the caGrid Query Language (CQL),
as shown in Figure 4. CQL is an object oriented query language that allows express-
ing queries with objects, related objects and attributes with desired values. While
the access interface to data is OO based, the data is kept in native formats (mostly
relational data or flat files). For the case of relational data caGrid provides the tools
for making object relational mappings and respective auto-generation of data access
software code.
1 OGSA-DAI [14] has been developed by the UK National eScience Centre and the
Universities of Edinburgh and Manchester. It allows exposing data resources on to grids,
including a collection of components for querying, transforming and delivering data, and
a simple toolkit for developing client applications.

104 P. Alper, C. Goble, O. Corcho
caBIG
Ontology
relational
data
O
G
SA
-
D
AI








CQ
L
Ac
tiv
itysemantic
aware client semantic query
Ontology
Service
access
SQL
convert
access
curator
O
n
to
lo
gy

to
o
ls Populate
manage
DAI service
descriptions
Metadata
Service
An
no
ta
tio
n
to
o
ls
curator
Populate
manage
Fig. 4. caGRID Data Access Services Architecture
4.5 Semantic OGSA-DAI
Semantic-OGSA-DAI [33] is an extension of OGSA-DAI for ontology-driven access
to relational data. It aims at supporting distributed query processing, semantic in-
tegration of distributed databases, and dynamic discovery of data sources depending
on their contents and capabilities.
In S-OGSA-DAI, metadata describes the relationships between relational data
sources and RDFS ontologies, using the D2R mapping language [26]. Metadata is
stored within S-OGSA-DAI and can be retrieved if needed. S-OGSA-DAI does not
impose a specific knowledge model to describe data sources, and assumes that the
data sources and the ontologies have been developed separately.
The extension of OGSA-DAI follows the extensibility option recommended for
OGSA-DAI. It defines a new activity that is able to process RDQL queries, as shown
in Figure 5. This approach is minimally intrusive, as the core of the OGSA-DAI
code remains unchanged and it does not affect the service interface. Instead, the
new functionality is developed as an add-on to the existing OGSA-DAI functionality.
For the implementation of this new activity, RDQL queries are translated to SQL
with the D2RQ [26] query translation engine.
4.6 Grid Meta-Scheduling Service
Meta-schedulers [6] are services that interface with multiple local schedulers (which
offer advance reservation of resources based on job execution start and stop times,
as well as at least partial access to local schedules) or other meta-schedulers to
negotiate with them advance reservation of resources based on user requirements,
such as time or QoS constraints. The goal of this negotiation is to determine feasible

Understanding Semantic Aware Grid Middleware for e-Science 105
mappings
ontology ? relationalDomain Ontology
relational
data
O
G
SA
-
D
AI








R
D
QL

Ac
tiv
ity
semantic aware
client
semantic query
Ontology Service
access
Metadata
SQL
convert
Fig. 5. Semantic OGSA-DAI Architecture
time slots in which all required resources are available for the requested start time
to execute a distributed workflow consisting of multiple jobs.
[7] proposes a semantic model to describe scheduler capabilities, where each local
scheduler or meta-scheduler contributes annotations for its capabilities to a common
metadata repository. Metadata is based on an OWL DL ontology, called the Grid
Scheduling Ontology (GSO), where each class represents a specific configuration
of scheduler capabilities. Schedulers are represented as individuals of one or more
classes of this ontology, hence reducing the scheduler selection problem to querying
the ontology model about all the known individuals of a class. The metadata is
relatively stable, given the slow rate of change of a scheduler’s capabilities.
This approach improves the accuracy of the scheduler pre-selection, and addi-
tional schedulers can be added to the pool at low cost. The new components, i.e.
those that represent an evolution with respect to [6], are shown in Figure 6. This
includes the use of ontology and reasoning services, a metadata repository and an
annotation service (the LS Semantic Adapter).
scheduler
capabilities
Grid Scheduling
Ontology
M
et
a
Sc
he
du
le
r
Metadata
Service
Client Reasoning
Submit Job to be executed
Local Scheduler
Submit
metadata
Ontology
Annotation Service
retrieve
Grid Scheduling
Ontology
retrieve
properties
convert
Ontology
Fig. 6. Meta-scheduler architecture

106 P. Alper, C. Goble, O. Corcho
4.7 Ontology and Role-Based VO Authorization
VO-AuthZ [18] is an ontology and role-based access-control service, which accepts
and produces XACML-compliant authorization requests and responses for a resource
access operation. Ontogrid-AuthZ calculates at run-time a subject’s eligibility to
access a resource, using a set of declaratively defined access control rules.
Metadata in VO-AuthZ describes domain attributes of the resource requestors
and is represented in RDF. The system does not impose a particular provisioning
process for metadata regarding subjects: this can be done manually or obtained
from other information sources. Metadata is stored and managed by the S-OGSA
Semantic Binding service [19].
Subject roles are defined with certain restrictions on subjects’ properties and
are deduced for a particular subject at run-time taking into account its properties
at that time and using instance reasoning (a. k. a. ABox reasoning). Role definitions
are based on the KAoS suite of ontologies [11], which contain descriptions of actors,
groups, actions, resources, policy types, etc. These ontologies are specialized to
describe domain specific access control policies and are implemented in OWL.
The system operation starts with the receipt of an authorization request, as
shown in Figure 7. The request contains the subject’s Distinguished Name and
further properties (i.e. RDF based metadata), which had been obtained by the caller
service by contacting known metadata servers (i.e. Semantic Binding services). This
information together with the VO ontology is passed onto a description logic classifier
(Pellet [25]), which performs instance level (A-box) reasoning to deduce the roles of
the subject. These roles are then used to decide whether access is allowed or not,
using a lookup table.
VO-AuthZ has been re-factored from an existing XACML-based service operat-
ing over access control lists for subjects. The old decision logic has been replaced
with the aforementioned one, while keeping the service interface intact. In the new
XACML request, information about the subject is submitted as attributes, including
additional RDF metadata about the subject, using XACML extensibility options. In
the absence of additional metadata the system would not be able to deduce subject
eligibility and would return an indeterminate response.
5 GRID MIDDLEWARE SYSTEM ANALYSIS
Tables 2, 3 and 4 summarise the results of applying the analysis framework to
the selected SAGMS. The information contained in the tables aims at giving some
light about their most characteristic features, the role of semantics in them and the
decisions taken in their development.
In our analysis, we identify three service groups: information services, data
access services and decision-support services. These groups are tightly related to
the type of OGSA service groups that the services belong to (information, data,
and the rest of groups that require more decision-support functions). However, the
borderlines between these groups are not rigid.

Understanding Semantic Aware Grid Middleware for e-Science 107
S
Y
S
T
E
M
S
em
a
n
ti
c
S
R
B
G
R
IM
O
IR
E
S
S
-M
D
S
ca
B
IG
D
a
ta
A
cc
es
s
S
er
v
ic
es
S
em
a
n
ti
c
O
G
S
A
-D
A
I
G
ri
d
M
et
a
S
ch
ed
u
li
n
g
V
O
A
u
th
o
-
ri
za
ti
o
n
Purpose
O
G
S
A
se
rv
ic
e
G
ro
u
p
In
fo
rm
at
io
n
In
fo
rm
at
io
n
In
fo
rm
at
io
n
D
at
a
D
at
a
E
xe
cu
ti
on
S
ec
u
ri
ty
F
u
n
ct
io
n
al
it
y
S
ea
rc
h
en
gi
n
e
R
eg
is
tr
y
In
d
ex
D
at
a
in
te
gr
at
io
n
D
at
a
in
te
gr
at
io
n
S
ch
ed
u
lin
g
R
ol
e-
b
as
ed
A
cc
es
s
co
nt
ro
l
A
d
d
ed
va
lu
e
fu
n
ct
io
n
al
it
y
N
ew
ca
p
ab
ili
ty
N
ew
ca
p
ab
ili
ty
N
ew
ca
p
ab
ili
ty
N
ew
ca
p
ab
ili
ty
N
ew
ca
p
ab
ili
ty
M
or
e
ac
cu
ra
cy
M
or
e
fl
ex
ib
ili
ty
Im
p
le
m
en
ta
ti
on
S
ta
n
d
ar
d
s
–
U
D
D
I
–
O
G
S
A
-D
A
I
O
G
S
A
-D
A
I
–
X
A
C
M
L
Development
S
of
tw
ar
e
S
ta
tu
s
P
ro
of
-O
f-
C
on
ce
p
t
P
ro
d
u
ct
io
n
P
ro
of
-O
f-
C
on
ce
p
t
P
ro
d
u
ct
io
n
A
lp
h
a,
P
ro
of
-
of
-C
on
ce
p
t
A
lp
h
a,
P
ro
of
-
of
-C
on
ce
p
t
A
lp
h
a,
P
ro
of
-
of
-C
on
ce
p
t
A
p
p
ro
ac
h
to
ob
ta
in
S
em
an
ti
c
A
w
ar
en
es
s
re
-w
or
ki
n
g
D
ev
el
op
m
en
t
fr
om
sc
ra
tc
h
re
-w
or
ki
n
g
re
-w
or
ki
n
g
re
-w
or
ki
n
g
re
-w
or
ki
n
g
re
-w
or
ki
n
g
R
e-
w
or
ki
n
g
M
et
h
od
A
u
gm
en
t
re
-w
ri
te
A
u
gm
en
t
A
u
gm
en
t
A
u
gm
en
t
A
u
gm
en
t
In
tr
u
si
ve
ex
te
n
si
on
E
xt
en
si
on
s
to
S
ta
n
d
ar
d
s
N
on
e
U
D
D
I
N
on
e
N
on
e
N
on
e
N
on
e
N
on
e
S
er
vi
ce
O
ri
en
ta
ti
on
N
o
Y
es
Y
es
Y
es
Y
es
Y
es
Y
es
Ta
bl
e
2.
G
rid
m
id
dl
ew
ar
ef
ea
tu
re
so
fs
ele
ct
ed
se
m
an
tic
aw
ar
e
gr
id
sy
st
em
s

108 P. Alper, C. Goble, O. Corcho
K
N
O
W
L
E
D
G
E
S
em
a
n
tic
S
R
B
G
R
IM
O
IR
E
S
S
-M
D
S
ca
B
IG
D
a
ta
A
ccess
S
erv
ices
S
em
a
n
tic
O
G
S
A
-D
A
I
G
rid
M
eta
S
ch
ed
u
lin
g
V
O
A
u
th
o
-
riza
tio
n
Entities
R
equ
ired
K
n
ow
-
led
ge
M
od
els
A
gn
ostic
A
gn
ostic
A
gn
ostic
(O
W
L
-S
by
d
efau
lt)
A
gn
ostic
A
gn
ostic
Y
es.
S
ch
ed
u
lin
g
ontology
Y
es.
R
ole
ontology
K
n
ow
led
ge
M
od
el
L
ifecycle
N
on
-frequ
ent
C
h
an
ges
N
on
-frequ
ent
ch
an
ges
N
on
-frequ
ent
C
h
an
ges
F
requ
ent
C
u
rated
C
h
an
ges
F
requ
ent
C
h
an
ges
N
on
-frequ
ent
ch
an
ges
F
requ
ent
C
h
an
ges
K
n
ow
led
ge
M
od
el
C
overage
D
om
ain
M
od
el
D
om
ain
M
od
el
D
om
ain
M
od
el
D
om
ain
M
od
el
D
om
ain
M
od
el
C
ontrol,
S
ch
ed
u
ler
p
re-selection
C
ontrol,
A
ccess
C
ontrol
R
u
les
K
n
ow
led
ge
M
od
el
G
en
eration
A
gn
ostic
A
gn
ostic
A
gn
ostic
A
gn
ostic
A
gn
ostic
A
gn
ostic
A
gn
ostic
D
ep
en
d
en
cy
level
to
K
n
ow
led
ge
O
p
tion
al
O
p
tion
al
O
p
tion
al
M
ission
C
ritical
M
ission
C
ritical
O
p
tion
al
M
ission
C
ritical
S
ecu
rity
N
o
secu
rity
N
o
secu
rity
N
o
secu
rity
S
ecu
red
ontol-
ogy
N
o
secu
rity
N
o
secu
rity
N
o
secu
rity
Technologies
K
n
ow
led
ge
R
ep
resentation
L
an
gu
ages
O
W
L
R
D
F
S
/O
W
L
O
W
L
U
M
L
,
O
W
L
R
D
F
(S
)
O
W
L
O
W
L
K
n
ow
led
ge
M
an
agem
ent
T
ools
an
d
In
frastru
ctu
res
Jen
a,
P
ellet
Jen
a,
S
esam
e
Jen
a
A
p
elon
O
ntology
S
erver
Jen
a,
S
esam
e,
D
2R
Q
W
S
-D
A
O
IN
T
,
P
ellet
O
W
L
-A
P
I,
P
ellet
K
n
ow
led
ge
M
od
el
A
ccess
local
N
/A
local
rem
ote
local
local
rem
ote
Table
3.K
nowledge
featuresofselected
sem
antic-aware
grid
m
iddleware
system
s

Understanding Semantic Aware Grid Middleware for e-Science 109
M
E
T
A
D
A
T
A
S
em
a
n
ti
c
S
R
B
G
R
IM
O
IR
E
S
S
-M
D
S
ca
B
IG
D
a
ta
A
cc
es
s
S
er
v
ic
es
S
em
a
n
ti
c
O
G
S
A
-D
A
I
G
ri
d
M
et
a
S
ch
ed
u
li
n
g
V
O
A
u
th
o
-
ri
za
ti
o
n
Entities
M
et
ad
at
a
b
as
ed
on
a
K
n
ow
le
d
ge
M
od
el
ye
s
n
o
ye
s
ye
s
ye
s
ye
s
ye
s
M
et
ad
at
a
L
if
ec
yc
le
F
re
qu
en
t
C
h
an
ge
s
F
re
qu
en
t
ch
an
ge
s
F
re
qu
en
t
C
h
an
ge
s
N
on
-f
re
qu
en
t
ch
an
ge
s
N
on
-f
re
qu
en
t
ch
an
ge
s
N
on
-f
re
qu
en
t
ch
an
ge
s
F
re
qu
en
t
C
h
an
ge
s
M
et
ad
at
a
C
ov
er
ag
e
D
om
ai
n
M
od
el
D
om
ai
n
M
od
el
D
om
ai
n
M
od
el
D
om
ai
n
M
od
el
an
d
T
ra
n
sf
or
m
at
io
n
D
om
ai
n
M
od
el
an
d
T
ra
n
sf
or
m
at
io
n
D
om
ai
n
M
od
el
D
om
ai
n
M
od
el
M
et
ad
at
a
G
en
er
at
io
n
M
an
u
al
A
gn
os
ti
c
A
u
to
m
at
ic
S
em
i-
au
to
m
at
ic
/
C
u
ra
te
d
M
an
u
al
M
an
u
al
fo
r
ca
p
ab
ili
ti
es
,
au
to
m
at
ic
fo
r
st
at
e
A
gn
os
ti
c
D
ep
en
d
en
cy
le
ve
l
to
M
et
ad
at
a
O
p
ti
on
al
O
p
ti
on
al
O
p
ti
on
al
M
is
si
on
C
ri
ti
ca
l
M
is
si
on
C
ri
ti
ca
l
O
p
ti
on
al
M
is
si
on
C
ri
ti
ca
l
A
cc
es
s
S
ec
u
ri
ty
N
o
se
cu
ri
ty
S
ec
u
re
d
N
o
se
cu
ri
ty
S
ec
u
re
d
N
o
se
cu
ri
ty
S
ec
u
re
d
N
o
se
cu
ri
ty
Technologies
M
et
ad
at
a
R
ep
re
se
nt
at
io
n
L
an
gu
ag
es
R
D
F
R
D
F
an
d
A
tt
ri
b
u
te
V
al
u
e
p
ai
rs
R
D
F
IS
O
/I
E
C
11
17
9
S
ta
n
d
ar
d
R
D
F
R
D
F
R
D
F
M
et
ad
at
a
M
an
ag
em
en
t
T
oo
ls
an
d
In
fr
as
tr
u
ct
u
re
s
Je
n
a
Je
n
a,
S
es
am
e
Je
n
a
O
ra
cl
e
R
D
B
M
S
Je
n
a,
S
es
am
e,
D
2R
Q
W
S
-D
A
O
IN
T
,
Je
n
a
O
nt
oG
ri
d
S
em
an
ti
c
B
in
d
-
in
g
S
er
vi
ce
M
et
ad
at
a
A
cc
es
s
lo
ca
l
lo
ca
l
lo
ca
l
re
m
ot
e
lo
ca
l
lo
ca
l
re
m
ot
e
Ta
bl
e
4.
M
et
ad
at
a
fea
tu
re
so
fs
ele
ct
ed
se
m
an
tic
-a
wa
re
gr
id
m
id
dl
ew
ar
e
se
rv
ice
s

110 P. Alper, C. Goble, O. Corcho
KaOS
ontology
Au
th
Z
Se
rv
ic
e
Ontology
ServiceMetadataService
Client Reasoning
authZ request + metadata
Metadata Provider
populate
retrieve
retrieve
Ontology
Fig. 7. Ontology and role-based VO Authorisation architecture
We make this classification to focus better on the features that are similar and
different in each group, so as to obtain better guidelines and lessons learned for future
developments. The distinguishing features of each service group are as follows:
• Semantic-aware Information Services use semantics to describe the metadata of
resources available in a system. Metadata is attached to those resources, stored
in repositories and harvested by the system clients mainly for resource discovery.
• Semantic-aware Data Access Services use metadata for the description of the
data sources to be accessed. Metadata consists in mappings between the data
source schema and a set of knowledge models, which have been normally develo-
ped separately. Mappings are executed when a system needs to access the data
and transform it into the knowledge model structure (e.g., by creating instances
of classes defined in them).
• Semantic-Aware Decision Support services use semantics to implement part or
all of their business logic. The services provide added-value functionality (more
accuracy, flexibility, etc.) than the ones provided without semantics. Normally
these services use the reasoning mechanisms associated to the knowledge models
used to represent metadata and the business logic.
In the following subsections we give more details about each group of services,
describing commonalities and differences between the systems described and pro-
viding guidelines and lessons learned for similar future developments.
5.1 Semantic-Aware Grid Information Services
Three services belong to this profile: S-SRB, GRIMOIRES and S-MDS. They can
be classified as OGSA information services, aimed at improving resource discovery.
Metadata is the major entity in this service group: services use ontology-based
metadata to describe existing resources, consolidate it in a common repository and
query it to discover resources. Hence metadata is treated as a first class citizen.
These services also provide metadata management facilities, such as lifetime man-
agement, change notification, etc., as described for S-MDS.

Understanding Semantic Aware Grid Middleware for e-Science 111
These systems do not impose any specific annotation process (that is, the process
of providing metadata for the resources that they refer to). They all assume that
metadata provision has been done either manually or by specialised systems (e.g.,
those able to transform the metadata already exposed by resources, normally in
the form of WS-Resource properties, into ontology-based metadata). In the case of
S-SRB, users can also provide this metadata manually with the extended mySRB
user interface. Only in S-MDS the metadata provisioning service is more coupled to
the internal architecture.
Each service is focused on annotation of different types of entities (files and
databases in S-SRB, services in GRIMOIRES and any type of resources in S-MDS).
To increase interoperation they recommend using a minimal generic metadata sche-
ma: UDDI and WSDL in GRIMOIRES, Dublin Core in S-SRB, and OWL-S in
S-MDS. However, the three systems are generic with respect to the domain knowledge
model used, in the sense that they could be easily applied to other types of Grid
entities by just changing the ontologies used for their descriptions.
With respect to the access to ontology-based metadata, GRIMOIRES and S-SRB
allow using non-semantic-aware clients (which will not benefit from the additional
services provided by them) as well as semantic-aware clients. S-MDS forces clients
to be semantic-aware in order to interact with the system. Semantic-aware clients
query these services in different ways:
• With API calls or system-specific objects (e.g. a set of attribute-value pairs),
which are matched against existing metadata. In this case the service is respon-
sible for calculating matches programmatically.
• With a query language (e.g. RDQL, SPARQL) suited to the underlying rep-
resentation formalism (RDF). In this case the query is built by the client and
forwarded to (and executed by) the underlying metadata storage system.
Currently these systems only support basic querying and some instance reason-
ing (as in S-SRB). Hence the awareness of specific domain knowledge and metadata
lies within the client applications, which interpret the results obtained from the ser-
vices. We think that, in the future, semantic-aware Grid services in this category
will provide more sophisticated matchmaking techniques (e.g. lexical similarity),
support for inexact/partial matches [10], etc.
5.2 Semantic Aware Grid Data Access Services
S-OGSA-DAI and the caGRID Data Services belong to this profile. They implement
services that can be classified inside the OGSA Data Access Service group.
Their common feature is that they all allow accessing distributed heterogeneous
data sources via the fac¸ade of a commonly agreed domain model. Here metadata
describes the schema-level mappings that can be established between data sources
and domain models. These mappings are executed when the data has to be retrieved
according to the domain model that the clients are able to deal with.

112 P. Alper, C. Goble, O. Corcho
The services are generic with respect to the domain knowledge models used.
These domain models can overlap completely with the data that is included in
the data sources, can be narrower or wider.
Another common feature of all these services is that they are implemented as
extensions of OGSA-DAI, which is one of the standards de facto for access to data
sources in Grid systems. As a result, the extensions done in both S-OGSA-DAI
and the caGrid Data Services are similar: they consist in the creation of a set of
activities for dealing with the format used to specify mappings and for executing
those mappings on the specified data sources. However, the interface that they
provide to clients is different in each case: S-OGSA-DAI offers an RDQL-aware
interface (clients have to specify their queries to the system in this language, which
will be translated to SQL) and the caGrid Data Services offer a proprietary CQL-
aware interface, an object-oriented query language that is suited to the specific type
of applications for which this extension is devised.
Similarly to the previous service category, a portion of semantic awareness also
lies mostly within the client applications of these services. These applications retrieve
and aggregate data views with respect to a common domain model, and they can
access ontology services and use ontology alignment and mapping techniques to
integrate data across multiple data grids, but this support is not provided by any of
the revised systems.
A difference between both systems lies in how they use the domain model.
S-OGSA-DAI uses the domain model as a schema that is populated by the data
transformed from the original data sources. In caGrid, data is transformed into an
object-oriented model and then it is strongly typed with elements from the domain
model.
In the future we foresee that similar semantic-aware data access services will
be developed, with a focus on new query languages and query result frameworks
and with the inclusion of heterogeneous data integration processes inside services,
as opposed to the current situation where clients perform their own integration. We
also foresee that the extensibility options of OGSA-DAI will still be used, due to its
robustness. Other foreseen extensions are related to the use of data access control
mechanisms, so that the sharing and access to data is better controlled.
5.3 Semantic-Aware Decision Support Grid Services
The two last services belong to this profile: the ontology and role-based VO Au-
thorisation system and the ontology-based Meta-Scheduler. The first one can be
classified as an OGSA Security Service and the second one as an OGSA Execution
Management Service, as shown in Table 2.
The common feature in this group of services is that they outsource a portion (or
all) of their business logic to an ontology and its associated reasoning mechanisms.
These services use application-domain ontologies (e.g., an application-specific exten-
sion of the KaOS ontology set in the VO AuthZ system and a scheduling ontology
in the meta-scheduler) to encode their business rules (e.g. rules of “who is autho-

Understanding Semantic Aware Grid Middleware for e-Science 113
rized” in KaOS). At run time, they take decisions based on the combination of the
metadata obtained from different resources and the use of the automated reasoning
mechanisms of the formalism that the business logic is represented with.
This contrasts with the conventional approach to implement control logic, which
mainly consists in hard-coding the system’s business rules into software programs.
The justification for using such type of approach in Grid systems is that business
rules can be, in many cases, complex, dynamic, and could be provided by multiple
parties. In such circumstances the conventional approach has the disadvantages of
1) rendering the business logic un-sharable, and 2) increasing code maintenance and
debugging costs.
The knowledge models used in this category are represented with richer knowl-
edge representation formalisms (e.g. restrictions expressed with axioms, rule lan-
guage constructs, etc.), due to the fact that they aim at capturing business logic.
With respect to metadata, annotation processes are normally application-depen-
dent. Besides, metadata can be deployed in different ways. It can reside alongside
knowledge, that is, the ontology driving the service business logic also contains
metadata describing the current situation of the system, as in the Meta Scheduler.
Or it can be hosted by metadata services and pushed-to or pulled-by the service for
its decision-making process, as in the VO AuthZ system.
Finally, the interfaces of these systems to their conventional clients are not
changed. Systems only provide an alternative way to perform an existing func-
tion inside the business logic of the system, with better quality of service properties
(more accuracy, flexibility, etc.).
We foresee that in the future other similar systems will be developed. Due to
the complex nature of the applications that can be developed using this approach, it
is not easy to characterise how they will be implemented. However, we foresee that
the main aspect of outsourcing the business logic of the system to external ontology
and reasoning services will be maintained.
6 SUMMARY
SAGMS are only recently making their way from research labs to real-life Grid
deployments. Hence this is a good time to start analysing the decisions taken in
their development and to compile the lessons learned, so that this can be taken into
account in the development of the next generation of systems. At the time of writing
this review we have not found another work that analyzes the role that knowledge
and metadata play in Grid middleware services.
In this paper we aimed at reviewing the main features of the early Grid middle-
ware systems that use semantics in e-Science applications. We focused on: 1) how
they have been developed and how they are used, 2) what their architectural features
are and 3) what is the role of knowledge and metadata in their architecture.
To improve the understanding of these services, we proposed an analysis frame-
work organised according to three dimensions: service development, metadata and

114 P. Alper, C. Goble, O. Corcho
knowledge. For each dimension we focused on a general description (WHAT) and on
how they are implemented (HOW). For each of them we identified several specific
features and we formulated questions to be answered or topics to be discussed.
We applied the framework to analyse in detail each of the seven Grid middleware
services selected. We classified these services into three categories: information ser-
vices, data access services and decision-support services. This classification allowed
us to focus on the characteristics of each system in the context of the problems
that they aim at solving, and to provide more insight about which are the typical
techniques and design proposals that are applied for each group. For example, we
have seen that those services devoted to data access use the extensibility options
of OGSA-DAI, that those devoted to information provisioning are focused on the
use of metadata to describe active resources and the consolidation of metadata in
common registries, and that those that provide decision-support reason with the
represented knowledge and metadata to take decisions.
Acknowledgements
This work is supported by the EU FP6 OntoGrid project (STREP 511513) funded
under the Grid-based Systems for solving complex problems, and by the Marie Curie
fellowship RSSGRID (FP6-2002-Mobility-5-006668).
REFERENCES
[1] Globus Toolkit: http://www.globus.org.
[2] EGEE gLite: http://glite.web.cern.ch/glite/.
[3] UNICORE: http://www.unicore.eu/.
[4] Open Middleware Infrastructure Institute: http://www.omii.ac.uk/.
[5] Miles, S.—Papay, J.—Payne, T.—Decker, K.—Moreau, L.: Towards a Pro-
tocol for the Attachment of Semantic Descriptions to Grid Services. In The Second
European across Grids Conference, Nicosia, Cyprus, p. 10, January 2004.
[6] Waldrich, O. et al.: A Meta-Scheduling Service for Co-Allocating Arbitrary
Types of Resources. In Proc. of the Second Grid Resource Management Workshop
(GRMWS-05) in conjunction with the Sixth International Conference on Parallel
Processing and Applied Mathematics.
[7] Missier, P.—Wieder, P.—Ziegler, W.: Knowledge and Data Management
in Grids. Volume 3 of CoreGRID, chapter Semantic Support For Meta-Scheduling
in Grids. Springer, 2006 (to appear).
[8] Carroll, J. J. et al.: Jena: Implementing the Semantic Web Recommendations.
Technical report, HP Labs, December 2003.
[9] Kawamura, T. et al.: Public Deployment of Semantic Service Matchmaker with
UDDI Business Registry. In S.A. McIlraith, D. Plexousakis, and F. van Harmelen,
editors, Proc. of the 2004 International Semantic Web Conference (ISWC 2004),
No. 3298 of Lecture Notes in Computer Science, pp. 471–485. Springer, 2004.

Understanding Semantic Aware Grid Middleware for e-Science 115
[10] Paolucci, M. et al.: Semantic Matching of Web Services Capabilities. In I. Horrocks
and J.A. Hendler, editors, First International Semantic Web Conference, Vol. 2342
of Lecture Notes in Computer Science, Springer, June 2002.
[11] Bradshaw, J. et al.: Representation and Reasoning for DAML-Based Policy and
Domain Services in KAoS and Nomads. In Proceedings of the Second International
Joint Conference on Autonomous Agents and Multiagent Systems, Melbourne, Aus-
tralia: ACM Press, 2003.
[12] Foster, I.—Kishimoto, H.—Savva, A.: The Open Grid Services Ar-
chitecture Version 1.0 Specification. Global Grid Forum OGSA Working
Group, 2005. https://forge.gridforum.org/projects/ogsa-wg/document/
draft-ggf-ogsa-spec/en/23.
[13] Czajkowski, K. et al.: Web Services Resource Framework (WSRF). Globus Alliance
and IBM, 2005.
[14] Antonioletti, M.—Atkinson, M.P.—Baxter, R.—Borley, A.—Chue
Hong, N.P.—Collins, B.—Hardman, N.—Hume, A.—Knox, A.—
Jackson, M.—Krause, A.—Laws, S.—Magowan, J.—Paton, N.W.—
Pearson, D.—Sugden, T.—Watson, P.—Westhead, M.: The Design and
Implementation of Grid Database Services in OGSA-DAI. Concurrency and Com-
putation: Practice and Experience, Vol. 17, 2005, Nos. 2–4, pp. 357–376.
[15] Saltz, J. et al.: caGrid: Design and Implementation of the Core Architecture of the
Cancer Biomedical Informatics Grid. Bioinformatics Journal, Vol. 22, 2006, No. 15,
pp. 1910–1916.
[16] Astakhov, V.—Gupta, A.—Grethe, J.—Ross, E.—Little, D.—
Yilmaz, A.—Martone, M.—Qian, X.—Santini, S.—Ellisman, M.:
Semantically Based Data Integration Environment for Biomedical Research.
Proceedings of the 19th IEEE International Symposium on Computer-Based Medical
Systems.
[17] Covitz, P.A.—Hartel, F.—Schaefer, C.—Coronado, S.—Fragoso, G.—
Sahni, H.—Gustafson, S.—Buetow, K.H.: caCORE: A Common Infrastructure
for Cancer Informatics. Bioinformatics Journal, Vol. 19, pp. 2404–2412.
[18] Alper, P. et al.: An Authorization Scenario for S-OGSA. Demo presentation at the
European Semantic Web Conference 2006.
[19] Corcho, O.—Alper, P.—Kotsiopoulos, I.—Missier, P.—Bechhofer, S.—
Goble, C.: An Overview of S-OGSA: A Reference Semantic Grid Architecture.
Journal of Web Semantics, Vol. 4, 2006, No. 2, pp. 102–115.
[20] Goble, C.—Corcho, O.—Alper, P.—de Roure, D.: e-Science and the Seman-
tic Web: A Symbiotic Relationship. In Proc. Discovery Science Conference 2006.
[21] Horrocks, I.—Patel-Schneider, P. F.: FaCT and DLP. In Proceedings of Auto-
mated Reasoning with Analytic Tableaux and Related Methods: International Con-
ference Tableaux-98, in Lecture Notes in Artificial Intelligence, pp. 27–30, Springer-
Verlag, May 1998.
[22] Friedman-Hill, E.: Jess Rule Engine for the Java Platform. Distributed Computing
Systems, 2003.
[23] Jastor Java code generator. http://jastor.sourceforge.net/.

116 P. Alper, C. Goble, O. Corcho
[24] Frey, J.G.: The CombeChem Experience. Knowledge Managment Workshop,
e-Science All Hands Meeting 2005, Notttingham, UK, September 19, 2005. UK,
EPSRC, http://eprints.soton.ac.uk/17452/.
[25] Grau, B.C.—Parsia, B.—Sirin, E.: Pellet, An OWL DL Reasoner. Proceedings
of the Third International Semantic Web 2004.
[26] Bizer, Ch.—Seaborne, A.: D2RQ – Treating Non-RDF Databases as Virtual
RDF Graphs. Poster, 3rd International Semantic Web Conference (ISWC2004).
[27] Horrocks, I. et al.: SWRL: A Semantic Web Rule Language Combining OWL
and RuleML. W3C Member Submission, May 21, 2004, http://www.w3.org/
Submission/2004/SUBM-SWRL-20040521/.
[28] Bellwood, T.: UDDI Version Specification. OASIS, July 2002, http://uddi.org/
pubs/uddi_v3.htm.
[29] Mirza Pahlevi, S.—Kojima, I.: S-MDS: A Semantic Information Service for Ad-
vanced Resource Discovery and Monitoring in WS-Resource Framework. GGF16 Se-
mantic Grid Workshop, Athens, February 2006.
[30] Dean, M.—Schreiber, G.: OWL Web Ontology Language Reference. W3C Re-
commendation, February 2004.
[31] Brickley, D.—Guha, R.—McBride, B.: RDF Vocabulary Description Language
1.0: RDF Schema. W3C Recommendation, February 2004.
[32] Klyne, G.—Carroll, J. J.: Resource Description Framework (RDF): Concepts
and Abstract Syntax. W3C Recommendation, February 2004.
[33] Kotsiopoulos, I.: Semantic OGSA-DAI. http://www.ontogrid.eu/s-ogsa-dai.
[34] Eres, M.H. et al.: Implementation and Utilisation of a Grid-Enabled Problem
Solving Environment in Matlab. Future Generation Computer Systems.
[35] Goble, C.A. et al.: Knowledge Integration: In Silico Experiments in Bioinformatics
in The Grid: Blueprint for a New Computing Infrastructure, Second Edition, eds.
Ian Foster and Carl Kesselman, 2003, Morgan Kaufman, November 2003.
[36] Sa´nchez-Gestido, M. et al.: Complex Data-intensive Systems and Semantic Grid:
Applications in Satellite Missions. In Proc. e-Science 2006 conference.
[37] Myers, T. S.—Atkinson, I.M.—Lavery, W. J.: The Semantic Reef: Managing
Complex Knowledge to Predict Coral Bleaching on the Great Barrier Reef. To appear
in: Proceedings of the AusGrid 2007 conference.
[38] Bachler, M.—Buckingham Shum, S.—Chen-Burger, Y.—Dalton, J.—
de Roure, D.—Eisenstadt, M.—Komzak, J.—Michaelides, D.—Page, K.—
Potter, S.—Shadbolt, N.—Tate, A.: Collaboration in the Semantic Grid:
A Basis for e-Learning. In Proceedings Grid Learning Services workshop, 7th Inter-
national Conference on Intelligent Tutoring Systems, 2004, pp. 1–12, Maceio, Brazil.
[39] Aldouri, R.—Keller, G. R.—Gates, A.—Rasillo, J.—Salayandia, L.—
Kreinovich, V.—Seeley, J.—Taylor, P.—Holloway, S.: GEON: Geophysical
Data Add the 3rd Dimension in Geospatial Studies. Proceedings of the ESRI Inter-
national User Conference 2004, Paper 1898; San Diego, California, August 90013,
2004.
[40] Science 2020. Microsoft Research, http://research.microsoft.com/
towards2020science/.

Understanding Semantic Aware Grid Middleware for e-Science 117
[41] de Roure, D.—Jennings, N.—Shadbolt, N.: Research Agenda for the Seman-
tic Grid: A Future e-Science Infrastructure. Technical report UKeS-2002-02, UK
e-Science Technical Report Series, National e-Science Centre, Edinburgh, UK. De-
cember 2001, http://www.semanticgrid.org/html/semgrid.html.
[42] Turk, Z.—Stankovski, V.—Gehre, A.—Katranuschkov, P.—
Kurowski, K.: InteliGrid Deliverable D13.1 – Semantic Grid Architecture.
2005.
[43] Snelling, D.—Fisher, M.—Basermann, A.: NextGrid Architecture White Pa-
per. December 2005, http://www.nextgrid.org/white_paper.htm.
[44] Jeffrey, S. J.—Hunter, J.: A Semantic Search Engine for the Storage Resource
Broker. GGF16 Semantic Grid Workshop, Athens, Greece. February 2006.
[45] Bray, T.—Paoli, J.—Sperberg-McQueen, C.M.—Maler, E.—Yergeau, F.
(Eds.): Extensible Markup Language (XML) 1.0. Fourth Edition. W3C Recommen-
dation. August 2006, http://www.w3.org/TR/XML/.
[46] Broekstra, J.—Kampman, A.—van Harmelen, F.: Sesame: A Generic Ar-
chitecture for Storing and Querying RDF and RDF Schema. International Semantic
Web Conference 2002, Sardinia, Italy.
[47] RDQL: http://jena.sourceforge.net/RDQL/.
Pinar is a research associate working within the Infor-
mation Management Group (IMG) at the School of Computer
Science of the University of Manchester. She currently partici-
pates in the EU FP6 IST project OntoGrid (FP6-511513), and
has participated in UK e-Science pilot project myGrid. Her re-
search work focuses on application of semantics in the areas of
Web Services and the service oriented Grid.
Oscar is a lecturer at the Ontological Engineering
Group at Universidad Politcnica de Madrid, and has worked
as a Marie Curie fellow at the Information Management Group
of the University of Manchester. His research activities include
the Semantic Grid, the Semantic Web and Ontological Engineer-
ing. Currently he participates at the EU FP6 IST project On-
toGrid (FP6-511513). He has published the books “Ontological
Engineering” and “A layered declarative approach to ontology
translation with knowledge preservation”, and over 30 journal,
conference and workshop papers.

118 P. Alper, C. Goble, O. Corcho
Carole is a professor of computer science at the Univer-
sity of Manchester and co-leads the IMG group since 1997. Her
research interests are on the accessibility of information, parti-
cularly the use of terminological and ontological services for the
representation and classification of metadata in a range of appli-
cation domains. Her recent work has been focused on two major
areas: the Semantic Web and e-Science/Grids; she has been in-
strumental in an effort to link the two areas by the application
of Semantic Web technologies to the Grid and e-Science, a fusion
dubbed by the Semantic Grid. She is co-chair of the Open Grid
Forum Semantic Grid Research Group.