130 likes | 252 Views
Architecture Scheduling Job Types. BQS Update. New Users Needs More users & machines, Scalability issues Needs for more sophisticated monitoring and control over the system. Yves.Fouilhe@in2p3.fr HEPIX Edinburgh 26/5/04. Architecture Update.
E N D
Architecture Scheduling Job Types BQS Update • New Users Needs • More users & machines, Scalability issues • Needs for more sophisticated monitoring and control over the system Yves.Fouilhe@in2p3.fr HEPIX Edinburgh 26/5/04
Architecture Update MySQL DB DB Agent query BQS Scheduler Client submit DB Agent results spawn report Worker Worker Yves.Fouilhe@in2p3.fr HEPIX Edinburgh 26/5/04
Scheduler Resources Quasi Interactive Jobs Scheduling Update Yves.Fouilhe@in2p3.fr HEPIX Edinburgh 26/5/04
More Control for Operation and Administration: Weight of Past Resource Usage And Group Objectives Max Job Duration Small Jobs Bias Scheduler Yves.Fouilhe@in2p3.fr HEPIX Edinburgh 26/5/04
Beyond Traditional Resources E.G. Disk, Time, Memory Logical Resources Name Max Available Restricted Flag Admin Defined Resources E.G. HPSS Logical Resources Yves.Fouilhe@in2p3.fr HEPIX Edinburgh 26/5/04
Created & managed by Users: Decide of the Name: u_XXX Receive Privilege bqs.u_XXXadmin Set Max Available and Restricted Flag Grant/deny bqs.u_XXXusage privilege Logical U_Resource Yves.Fouilhe@in2p3.fr HEPIX Edinburgh 26/5/04
A General Service, APIs, Commands To: Grant, Deny, Check & List Privileges Given to Users, Groups and Machines EG in BQS applid: bqs.admin, bqs.oper, bqs.spawn_forbidden Privilege Management Yves.Fouilhe@in2p3.fr HEPIX Edinburgh 26/5/04
Parallel Jobs Arborescent Jobs GRID Jobs New Job Types Yves.Fouilhe@in2p3.fr HEPIX Edinburgh 26/5/04
2 new submit options: proc, ptype proc: Number of WorkPoints ptype: PVM, MPICH, LAM-MPI Parallel Jobs Yves.Fouilhe@in2p3.fr HEPIX Edinburgh 26/5/04
Parallel Jobs query MySQLDB BQS DB Agent Client submit … BQS Master DB DB Agent results spawn parallel job global report spawn task Worker Worker report Yves.Fouilhe@in2p3.fr HEPIX Edinburgh 26/5/04
SNOVAE: many related small tasks need short global response time Schedule and spawn as one Job to reduce BQS latency Runs on a number of WorkPoints User must describe tasks dependencies Arborescent Job Yves.Fouilhe@in2p3.fr HEPIX Edinburgh 26/5/04
Real and Generic Accounts AFS Tokens and Certificates Specific RH 7.3 + LCG Soft Full Production Farm (Currently a Specific “lcg” Logical Test Farm for Validation) GRID Jobs Yves.Fouilhe@in2p3.fr HEPIX Edinburgh 26/5/04
BIO: Quasi Interactives Jobs Installation and documentation for LCG Other Projects Yves.Fouilhe@in2p3.fr HEPIX Edinburgh 26/5/04