Resources on a computational Grid are geographically istributed,heterogeneous in nature,owned by different individuals or organizations with their own scheduling policies,have different access cost models with dynamically varying loads and availability conditions.This maker traditional approaches to workload management,load balancing and scheduling inappropriate.The first work package(WP1)of the EU-funded DataGrid project is adddressing the issue of optimzing the distribution of jobs onto Grid resources based on a knowledge of the status and characteristics of these resources that is necessarily out-of -date(collected in a finite amonut of time at a very loosely coupled site).We describe the DataGrid approach in integrating existing software components(from Condor,GGlobus,etc.)to build a Grid Resource Broker,and the early efforts to define a workable scheduling strategy.
展开▼
