Agent Based Resource Management in Computational Grids

Omer Rana
Omer's Internet Page

Friday, 27th July 2001
0930 - 1100

Computer Science, RMIT

Abstract:

"The ``Grid'' is an emerging infrastructure that connects multiple regional and national resources to create a universal source of computing power -- the word ``Grid'' was chosen by analogy with the [electric] power grid, which provides pervasive access to electrical power. We believe that by providing pervasive, dependable, consistent and inexpensive access to advanced computational capabilities, databases, sensors, and people, computational grids will have a transforming effect similar to the electric power grid, allowing new classes of applications to emerge." (Ian Foster and Carl Kesselman [Foster99])
Computational Grids are an important new research area which combine aspects of applications and computer science. The Computational Grid community draws its analogies from the electricity (power) Grid, whereby computational resources can be shared based on demand. I present two themes in this talk to enable the agent based paradigm to support resource management in discovery in Computational Grids.

1. How can an agent based design paradigm support the establishment of Grid environments.

A de-centralised approach to resource management and discovery, based on a community of interacting software agents is described. Each agent either represents a user application, a resource, or a MatchMaking service. The proposed approach can support dynamic registration of resources and user tasks, facilitating the establishment of dynamic clusters. Resource capability and task requirements are described using an object based data model, enabling new types of devices or new features in existing devices to be identified. A comparison with the Discovery and LookUp services in Jini and TSpaces is also provided.

2. Supporting a large number of resource -- what are the issues in supporting Scalability in the system outlined in theme 1

Scalability is an issue that becomes important when developing practical software agent systems, or when comparing infrastructure to implement multi-agent systems. Scalability is crucial in the context of Computational Grids, where agents may wrap or manage different kinds of resources, with different characteristics. It is important for an application developer, for instance, to be able to identify aspects of an agent system that may prevent scaling beyond a certain number of agents or users. Scalability is also a term that has been applied in different contexts in the agents community, and due to diversity of approaches in the field, can imply different things. Multi-agent scalability must take into account themes which are specific to agent systems, such as co-ordination and communication between agents, behaviour of each agent, and higher level structures such as agent organisation, and the emergence of such structures through collective behaviours of participants. It is also important to be able to measure and compare agent systems, for specific applications, to be able to dete rmine whether one of these has attributes which support scalability. I outline scalability metrics that may be used to compare two agent designs to identify which one could be `more' scalable than another.

Biosketch:

Omer Rana is a lecturer in computer science at Cardiff University, and currently visiting Lin Padgham, James Harland and Michael Winikoff at RMIT. He holds a PhD from Imperial College (London University), and has worked in projects at Argonne National Laboratory (US), Oak Ridge National Laboratory (US), BT Labs, and the University of Syracuse (US). He is also actively involved in Grid activities via the Global Grid Forum where he co-chairs a working group, and is co-founder of the regional Grid centre at Cardiff University (one of 8 in the UK). His research interests are in high performance distributed computing and multi-agent systems.

REFERENCES: [foster99] I. Foster and C. Kesselman, The Grid: Blueprint for a New Computing Infrastructure Morgan Kaufmann Publishers, 1999

More details of theme 1 can be found in: ``Resource Discovery for Dynamic Clusters in Computational Grids'' Omer F. Rana, Daniel Bunford-Jones, David W. Walker, Matthew Addis, Mike Surridge and Ken Hawick Proceedings of ``Heterogeneous Computing Workshop'' at IPDPS/SPDP, San Francisco, California (April 2001).

More details of theme 2 can be found in: ``What We Mean When We Talk About Multi-Agent Scalability'' Omer F. Rana and Kate Stout, Proceedings of Autonomous Agents Conference 2000, Barcelona, Spain. ACM Press