The CMS experiment at CERN is setting up a Grid infrastructur required to fulfill the needs imposed byTerabyte scale productions for the next few years. The functionality envisaged by this one-year project spans the Distributed Data Management aspects of the Grid, in near future, and the Distributed Computing aspects for the later phases. The goal is to automate the production as much as possible but at the same time allow the users to interact with the system, if required, to make decisions which would optimize performance.
We present the architecture, design and functionality of our first working Data Management prototype. The middle ware of choice is the Globus toolkit that provides promising functionality. We present test results which prove ability of the Globus Toolkit to be used as an underlying technology for a world-wide Data Grid. The required Data Management functionality includes, high speed file transfers, secure access to remote files, selection and synchronization of replicas and managing the meta information. The whole system is expected to be flexible enough to incorporate site specific policies. The Data Management granularity is the file rather than the object level. The second phase of the project will cover developing Distributed Computing tools that provide global scheduling algorithms and uniform interfaces with the heterogeneous local schedulers. The first prototype is currently being tested and will be in place for the High Level Trigger (HLT) production in autumn 2000. Owing to these efforts, CMS will be one of the pioneers to use the Data Grid functionality in a running production system with data volumes ranging anywhere from tens to hundereds of Terabytes. The project can be viewed as an evaluator of different strategies, a test for the capabilities of middle ware tools and a provider of basic Grid functionalities.