Programming Research Group Research Report RR-04-26

Lee Momtahan and Andrew Martin

February 2004, 19pp.

Abstract

DataGrids seek to pool geographically disperse storage and computational resources to solve large-scale data-intensive problems. The EU-DataGrid is a large research project attempting to build one of the first such DataGrids. Effective job submission and scheduling in DataGrids is a significant challenge. We present an account of the EU-DataGrid scheduling architecture, some typical usage scenarios, and a discussion of some of the issues that arose when this architecture was used in practice. Some of these issues are generic, and likely to be of interest to future attempts to build a DataGrid, but also of a pragmatic nature, illuminated by experience as well as theory.