1. Introduction
Grid computing [4] has acquired immense popularity as a platform which enables collaborations amongst scientists in the form of data sharing and remote processing [6]. The overall end-to-end quality-of-service of an application is dependent on reliable performance [3] obtained from each component of Grid middleware. The Grid Information Service is a core component since it enables the discovery and selection of resource entities on the Grid. Moreover, Grid environments create the implicit need for applications to obtain real-time information about the structure and state of the meta-system, to utilise this information to make configuration decisions, and to be notified when information changes. Moreover, the heterogeneity, dynamism and complexity of the Grid infrastructure can be tackled by the emergence of a relatively new programming paradigm and management technique called autonomic computing [7]. An autonomic computing system is one which can self-manage, self-define, self-configure and self-optimise. This paper addresses these issues whilst drawing on the popular Grid Information Service, the OGSA-based (Open Grid Services Architecture) [5] MDS3. A new approach is proposed, using the dynamic adjustment of the notification rate for preventing the GIS from overloading and improving its service performance. This is done by characterising and evaluating the performance achievable by the Globus [2] Monitoring and Discovery System (MDS3) [1] which is a widely deployed reference implementation of a Grid information service. The context of this paper is therefore the autonomic delivery of dynamic, up-to-date events to clients using the MDS3 push-based mechanism.