The Active Information Repository

In this paper, we describe the design of a scalable grid node to support computation on information stored in databases. It is envisaged that huge amounts of information, both structured and semi-structured, will be made available on The Grid. Simple client-server style access to the information server will be inadequate, as the network connecting the client to the server will not have the required bandwidth to support computations that may access more than Petabytes of data. A solution that allows clients to submit computations to be run locally to the data is therefore required. The Active Information Repository is designed for this. Data is stored in a scalable object database server and computations, in the form of agents, can be sent by clients for execution on a scalable agent execution server that is tightly coupled, through a high bandwidth network, to the database server. The paper describes the rationale for this approach, and describes the design of the major components: the parallel database server, and the parallel agent execution server.