ResourcesonacomputationalGridaregeographicallyistributed,heterogeneousinnature,ownedbydifferentinpidualsororganizationswiththeirownschedulingpolicies,havedifferentaccesscostmodelswithdynamicallyvaryingloadsandavailabilityconditions.Thismakertraditionalapproachestoworkloadmanagement,loadbalancingandschedulinginappropriate.Thefirstworkpackage(WP1)oftheEU-fundedDataGridprojectisadddressingtheissueofoptimzingthedistributionofjobsontoGridresourcesbasedonaknowledgeofthestatusandcharacteristicsoftheseresourcesthatisnecessarilyout-of-date(collectedinafiniteamonutoftimeataverylooselycoupledsite).WedescribetheDataGridapproachinintegratingexistingsoftwarecomponents(fromCondor,GGlobus,etc.)tobuildaGridResourceBroker,andtheearlyeffortstodefineaworkableschedulingstrategy.
