Simulation on the GRID lesson learned from Ba

Simulation on the GRID: lesson learned from Ba. Bar Eleonora Luppi Ferrara University & INFN E. Luppi Super. B Computing Workshop - Frascati 17 -12 -2008 Super. B Computing Workshop Frascati 17 -12 -08

Overview • • • Goal Ba. Bar SPGrid summary Ba. Bar vs Super. B What we need to avoid in Super. B Conclusions Note : personal recollections of lesson learned in Ba. Bar E. Luppi Super. B Computing Workshop Frascati 17 -12 -08 2

What can we ask a Grid Monte Carlo Production System for? Our goal is a production system with: • Single production manager defining jobs • • • Automatic job creation and submission Worldwide Grid of remote sites Automatic resource allocation Real-time job monitoring Automatic output retrieval and cataloguing Dynamic failure detection and recovery E. Luppi Super. B Computing Workshop Frascati 17 -12 -08 3

Ba. Bar SPGrid - Integration We had to manage the Ba. Bar production tools integration with Grid • Prod. Tools – spsub integrated with standard Grid submission – Output retrieved from SE to UI using a custom script – spmerge integrated in standard Grid submission • “Official” SP release import (dependencies!) • New Grid-compliant software package E. Luppi Super. B Computing Workshop Frascati 17 -12 -08 4

Ba. Bar SPGrid - Installation The (grid version) Ba. Bar Simulation software is packaged in a ~70 MB tar ball. It includes: • Monte Carlo simulation binaries • Merging tools to manage job output simulated events • Export tools for transferring the simulated events file to SLAC and update the central bookkeeping DB The software is installed on INFN-GRID sites CE by standard LCG procedure • Selected sites for SP tagged for job submission • Requirements specified into job’s JDL file E. Luppi Super. B Computing Workshop Frascati 17 -12 -08 5

Ba. Bar SPGrid - Resources Ba. Bar SP jobs access different data input types: • Detector background • Detector conditions • Calibration constants Input data are stored as ROOT I/O files in 4 INFNGRID sites (CNAF, Ferrara, Padova, Pisa) ROOT Input data file accessed “off-grid” by Xrootd server facility, locally or via Wide Area Network Data file info via General Parallel File System (GPFS) E. Luppi Super. B Computing Workshop Frascati 17 -12 -08 6

Ba. Bar SPGrid - workflow • Jobs submitted from the User Interface (CNAF) to a dedicated Resource Broker (Ferrara) • Workload distributed by the RB over Computing Elements of all involved sites • Jobs access Root input data from Xrootd servers • Jobs output registered into LFC catalogue and stored to Closest Storage Elements for later retrieving • On UI the monitor system daemon checks the jobs status, retrieves the jobs output, and applies submission retry and recovery policies • On UI the simulated event files are merged and exported to SLAC E. Luppi Super. B Computing Workshop Frascati 17 -12 -08 7

Ba. Bar vs Super. B • Ba. Bar was born before Grid, so we are still managing a hybrid infrastructure to run Monte Carlo on Grid – – – Data access and storage Bookkeeping “Special” grid package Releases validation on variable platforms No end user interaction • Super. B is a new project, so we must set-up a Simulation Production system fully compatible with Grid infrastructure E. Luppi Super. B Computing Workshop Frascati 17 -12 -08 8

Warning … Dependencies management is crucial: • • • Proper abstraction encapsulation use of generic programming clear separation of software into packages attention to forward and backward compatibility attention to external software: – external software coupled to Ba. Bar software (at compile, link time) has complicated code development, restricted our choice of platforms and resulted in painful migrations E. Luppi Super. B Computing Workshop Frascati 17 -12 -08 9

Conclusions We can start to organize a Monte Carlo production on Grid developing a Grid-compliant Simulation software We need to take care of the Ba. Bar lesson and other (LHC i. e. ) experiments experience Not only a “grid team” has to be involved, but also end users and code developers • tutorial This good start could be useful to go toward a Grid analysis environment E. Luppi Super. B Computing Workshop Frascati 17 -12 -08 10

Ba. Bar SPGrid - workflow diagram E. Luppi Super. B Computing Workshop Frascati 17 -12 -08 11