WP 1 progress report Cristina Vistoli Massimo Sgaravatto

WP 1 – progress report Cristina Vistoli, Massimo Sgaravatto, Francesco Prelz, Antonia Ghiselli, Cosimo Anglano, Luciano Gaido, Franco Semeria, Salvatore Cavalieri For Application Req. : Ivano Lippi, Claudio Grandi, Lucia Silvestris 4/9/2000 I Workshop Datagrid C. Vistoli

Progress Report 2 INFN weeks with Globus Team members Steve Tuecke, Lee Liming + Miron Livny (Condor Team) Ann Cheverniak, Steve Fitzgerald Workload management applied to real HEP application: Event production CMS-HLT production analyzed in terms of GRID requirements and Globus tools availability Simplified description: 4/9/2000 I Workshop Datagrid C. Vistoli

HLT production simplified description PD I 4/9/2000 O BO BA… O O I Workshop Datagrid C. Vistoli CERN O

Application Grid Requirements The GRID requirements are: Resource Broker to select, find allocate the computing resources where the jobs have to be submitted. move the input files in the selected farm and submit the jobs collect the output files produced in each farm in a central repository 4/9/2000 I Workshop Datagrid C. Vistoli

Resource Broker Architecture Job code (class-ads) Master condor_submit LDAP to Class-ads converter condor-rm Grid Information Server (LDAP Server) log Condor schedd Globus submit GRAM Job Manager LSF PC Cluster 4/9/2000 PBS PC Cluster Condor PC Cluster I Workshop Datagrid C. Vistoli

Comments: Condor Scheduler is a robust tool Class-ads are flexible enough to describe the various job policies Job submission from one/more location to the whole GRAM resources GRAM/Job Manager to submit Job to the selected Farms, using Globus Security 4/9/2000 I Workshop Datagrid C. Vistoli

Work needed GRAM: to provide all the necessary info to the Grid Information System Collaboration with Globus team on-going: Massimo Sgaravatto presentation Need to define “what” is the necessary info in PC cluster resource environment GRAM/job manager persistent and fault tolerant Massimo Sgaravatto presentation Export the information about computing systems availability, disk and network from the LDAP server to the Class-ad collection 4/9/2000 I Workshop Datagrid C. Vistoli

Work needed - cont Master implements job policy and monitor the job execution. A real possibility to verify if match-making algorithm satisfy application requirements. Master access Class-ads collection to select the GRAM resource Input files are moved to the selected resource using ‘gsiftp’ using third part transfer. Input files are normally small, no need for advance network reservation 4/9/2000 I Workshop Datagrid C. Vistoli

Work needed - cont To complete the application execution: Output files have to be collected in the central repository Data Mover and Replica Manager are useful tools. Strict collaboration with WP 2 Investigate the use of advance network reservation for these files transfer 4/9/2000 I Workshop Datagrid C. Vistoli

Documentation www. infn. it/grid 4/9/2000 I Workshop Datagrid C. Vistoli