Semantic-Based, Scalable, Decentralized and Dynamic Resource Discovery for Internet-Based Distributed System

Resource Discovery (RD) is a key issue in Internet-based distributed sytems such as
grid. RD is about locating an appropriate resource/service type that matches the user's
application requirements. This is very important, as resource reservation and task
scheduling are based on it. Unfortunately, RD in grid is very challenging as resources
and users are distributed, resources are heterogeneous in their platforms, status of the
resources is dynamic (resources can join or leave the system without any prior notice)
and most recently the introduction of a new type of grid called intergrid (grid of grids)
with the use of multi middlewares. Such situation requires an RD system that has rich
interoperability, scalability, decentralization and dynamism features. However,
existing grid RD systems have difficulties to attain these features. Not only that, they
lack the review and evaluation studies, which may highlight the gap in achieving the
required features. Therefore, this work discusses the problem associated with intergrid
RD from two perspectives. First, reviewing and classifying the current grid RD
systems in such a way that may be useful for discussing and comparing them. Second,
propose a novel RD framework that has the aforementioned required RD features. In
the former, we mainly focus on the studies that aim to achieve interoperability in the
first place, which are known as RD systems that use semantic information (semantic
technology). In particular, we classify such systems based on their qualitative use of
the semantic information. We evaluate the classified studies based on their degree of
accomplishment of interoperability and the other RD requirements, and draw the
future research direction of this field. Meanwhile in the latter, we name the new
framework as semantic-based scalable decentralized dynamic RD. The framework
further contains two main components which are service description, and service
registration and discovery models. The earlier consists of a set of ontologies and
services. Ontologies are used as a data model for service description, whereas the
services are to accomplish the description process. The service registration is also based on ontology, where nodes of the service (service providers) are classified to
some classes according to the ontology concepts, which means each class represents a
concept in the ontology. Each class has a head, which is elected among its own class
I
nodes/members. Head plays the role of a registry in its class and communicates with
I
the other heads of the classes in a peer to peer manner during the discovery process.
We further introduce two intelligent agents to automate the discovery process which
are Request Agent (RA) and Description Agent (DA). Eaclj. node is supposed to have
both agents. DA describes the service capabilities based on the ontology, and RA
I
carries the service requests based on the ontology as well. We design a service search
I
algorithm for the RA that starts the service look up from the class of request origin
first, then to the other classes.
We finally evaluate the performance of our framework ~ith extensive simulation
experiments, the result of which confirms the effectiveness of the proposed system in
satisfying the required RD features (interoperability, scalability, decentralization and
dynamism). In short, our main contributions are outlined new key taxonomy for the
semantic-based grid RD studies; an interoperable semantic description RD component
model for intergrid services metadata representation; a semantic distributed registry
architecture for indexing service metadata; and an agent-qased service search and
selection algorithm.
Vlll