In recent years the case for the integration of agent-based and grid-based technologies has grown ever more compelling, particularly with the growth in business data volumes, the heterogeneity of data and information sources, the increasing need for interoperability between systems and the emergence of practical grid computing applications.
The majority of current research effort in this area is focussed at using multi-agent systems to improve grid middleware by providing enhanced scheduling, fault tolerance or self-organising capabilities.
The G-MAS (Grid Enabled Multi-agent Systems) project proposes to investigate how existing grid resources can be harnessed by agents to reduce computational demands on the underlying agent platform. Scalability is seen by many from both within and outside the agent research community as an issue that has the potential to limit the real-world take-up of such systems.
Grid computing has seen a period of sustained growth and is gradually becoming a widely adopted methodology for the construction of high performance computational facilities. With many established grid platforms already in operation the project will explore how these resources can be used without modification to enhance the scalability of the SAGE platform.
SAGE is a distributed decentralized, fault tolerant, scalable and lightweight agent platform that has been designed with the aim of maximising the scalability and stability of the multi-agent systems that run on it. A relatively new addition to the SAGE architecture is the ability to call upon Globus based grid-services. This is accomplished using a semantic translation between the agent infrastructure and web services that provide access to the underlying grid resources.
The semantic knowledge used during the translation process is provided in an ontology which contains a semantically rich description of the Grid resources that are available in the local environment. This allows agents to find suitable facilities for a given purpose.
At present, the SAGE architecture has limited access methods for communication with grid middleware and hence only Globus-based grid services can be used. A wide-range of middlewares are currently deployed in production quality grids all over the world, such as those at CERN which use a g-lite based infrastructure for handling data from experiments. Therefore it could be useful to investigate how such resources can be opened up to systems running on a SAGE platform.
The G-MAS project
The G-MAS project aims to build on the work that has already started within the SAGE project regarding the integration of multi-agent systems, web services and grid middleware. The focus will be on improving the scalability of the platform using grid-based methods to reduce the complexity of individual agents. This process will begin with a thorough analysis of the resources that different grid middleware provides.
Following this stage, generic services will be identified and integrated within the architecture of SAGE platform. Ontological representation will once again be key to this aspect of the system, providing the necessary description of grid capabilities and methods of interaction.
UWE has developed expertise in the design and use of ontological representations and their access using standard languages such as OWL, especially in the context of distributed medical data mining on grid infrastructures.
As an example of a computationally demanding process that can be run on the grid instead of internally within an agent, test data mining tasks will be created that use different grid resources. This will enable researchers to evaluate the efficiency of the new structures and to perform comparative studies with existing agent platforms such as JADE. It is expected that the work pioneered in G-MAS will lead to strong and fruitful collaboration between UWE and NIIT in the area of agent-based grid data mining.