Download presentation
Presentation is loading. Please wait.
Published byJerome Blake Modified over 9 years ago
1
The ALICE short-term use case DataGrid WP6 Meeting Milano, 11 Dec 2000Piergiorgio Cerello 1 Physics Performance Report (PPR) production starting in Feb2001 Opportunity for a "zero-order" test with GRID tools Get many people in touch with GLOBUS Be ready to make use of newly available GRID tools Better define our requirements "experimentally" Improve & simplify the production control?
2
The ALICE PPR DataGrid WP6 Meeting Milano, 11 Dec 2000Piergiorgio Cerello 2 Evaluate the ALICE performance with respect to the detection of a set of interesting physics signals Pb-Pb event generation (HIJING, driven by AliRoot) Event tracking (GEANT3, driven by AliRoot) Full detector simulation (AliRoot) List of "Digits" (Raw Data) Step 1: background simulation
3
The ALICE PPR DataGrid WP6 Meeting Milano, 11 Dec 2000Piergiorgio Cerello 3 Physics signal generation & digitization Background superposition Reconstruction List of Points, Tracks, Vertices Step 2: signal simulation + reconstruction Evaluation of: - signal detection acceptance & efficiency - mass resolution Step 3: Analysis
4
Simulation DataGrid WP6 Meeting Milano, 11 Dec 2000Piergiorgio Cerello 4 Requirements on mass storage & CPU mainly come from the simulation of "background" events Requirements on mass storage & CPU mainly come from the simulation of "background" events About 10 4 background events needed (8.4 x10 4 primary tracks each) Mass storage: about 1 GB/event > 10 TB storage needed Step 1 CPU: about 24 hours/event on PIII 600 Mhz, 512 MB RAM Memory: about 450 MB/job
5
Simulation DataGrid WP6 Meeting Milano, 11 Dec 2000Piergiorgio Cerello 5 node:/path/run#/event#/galice.root (1 GB) node:/path/run#/event#/ _sumDigits.root (100 MB) galice.root ----> Header, Kine, Hits _sumDigits.root ----> Digits for detector = ITS, TPC, MUON, PMD, RICH, ZDC & so on Data output structure: 1 event/directory Input Config.C grun.C job_id.err job_id.dat Output
6
Signal + Reconstruction DataGrid WP6 Meeting Milano, 11 Dec 2000Piergiorgio Cerello 6 Remarks Step 2 Requirements on workload management & data access Requirements on workload management & data access Organization will depend on the total input size - Need to make input data (background digitized events) available to any user - Need to superimpose different physics signals on the same background event
7
Signal + Reconstruction DataGrid WP6 Meeting Milano, 11 Dec 2000Piergiorgio Cerello 7 Step 2 - Option 1 - All the digitized data stored at a single site - Access to any user + data shipping to the user's site -> processing - Reconstruction output stored at the user's site - All the digitized data stored at a single site - Access to any user + data shipping to the user's site -> processing - Reconstruction output stored at the user's site Remarks: - Need for a Workload manager - Network performance: 1 TB x # of users - Need for a Workload manager - Network performance: 1 TB x # of users
8
Signal + Reconstruction DataGrid WP6 Meeting Milano, 11 Dec 2000Piergiorgio Cerello 8 Step 2 - Option 2 - Distributed storage of digitized data - Scheduling of signal superposition & reconstrution jobs - Reconstruction output stored at the user's site - Distributed storage of digitized data - Scheduling of signal superposition & reconstrution jobs - Reconstruction output stored at the user's site Remarks: - Workload managed locally - Network performance: 0.1 TB x # of signals - Workload managed locally - Network performance: 0.1 TB x # of signals
9
Signal + Reconstruction DataGrid WP6 Meeting Milano, 11 Dec 2000Piergiorgio Cerello 9 Step 2 - Option 3 - Distributed storage of digitized data - Access to any remote user + data shipping to the user's site - Reconstruction output shipped to a single site (which might be different for different signals) - Distributed storage of digitized data - Access to any remote user + data shipping to the user's site - Reconstruction output shipped to a single site (which might be different for different signals) Remarks: - Need for a Workload manager - Network performance: 1 TB x # of users - Need for a Workload manager - Network performance: 1 TB x # of users
10
Analysis DataGrid WP6 Meeting Milano, 11 Dec 2000Piergiorgio Cerello 10 Step 3 - Distributed storage of reconstructed data - Access to any remote user + data shipping to the user's site - Analysis - Distributed storage of reconstructed data - Access to any remote user + data shipping to the user's site - Analysis Remarks: - Workload managed locally - Need for a data manager - Network performance: 0.1 TB x # of users - Workload managed locally - Need for a data manager - Network performance: 0.1 TB x # of users
11
The ALICE PPR: where? DataGrid WP6 Meeting Milano, 11 Dec 2000Piergiorgio Cerello 11 Sites quoted for the PPR production No detailed sharing decided yet CERN Amsterdam - NIKHEF Cagliari - INFN Catania - INFN Columbus - Ohio State University Darmstadt - GSI Dubna - JINR Lyon - IN2P3 Padova - INFN Torino - INFN
12
Status: ALICE Software Milano, 11 Dec 2000Piergiorgio Cerello GCC 2.95.2 as compiler, recompiled for i686 ROOT v 2.25/03 patched & compiled AliRoot v3.04pre AliRoot is rapidly evolving towards version 3.04 which will be used for the Physics Performance Report production AliRoot framework development for the PPR production at CERN GRID (Globus) tests between sites quoted for the PPR production installed and working at all sites - preliminary installation kit available DataGrid WP6 Meeting 12
13
Status: GRID Software Globus v.1.1.3 installed in Cagliari, Catania, Columbus, Lyon, Torino Dubna, GSI, Saclay: expected by the end of the year Merida: work just starting on the GRID item Milano, 11 Dec 2000Piergiorgio Cerello 13 DataGrid WP6 Meeting Lower-level job manager: Cagliari, Catania, GSI: LSF Lyon: BQS Torino: PBS Columbus, Dubna, GSI, Merida, Saclay: to be defined
14
Test activities - up to now overhead tests using a CPU-bound program; benchmark tests using NAS (System Division at NASA Ames Research Center) Software functionality tests (elementary job submission using globus-job-run and globusrun commands); Run AliRoot (1 full event) with the whole Alice detector on the farms [no Grid] test submission and execution stages on the PBS (LSF,...) farm driven by a local host [Grid] Milano, 11 Dec 2000 14 DataGrid WP6 Meeting Piergiorgio Cerello
15
Test activities - up to now driven by GLOBUS submitted from Torino, executed in Catania CPU load monitored through the WEB data output stored remotely stdout & stderr shipped back to submitting machine Execution time: 24 h Milano, 11 Dec 2000 15 DataGrid WP6 Meeting Piergiorgio Cerello Remote generation of a full ALICE background event
16
Test activities - plans test submission and execution stages on local farms driven by remote sites [Grid] repeat between different partners [Grid] Milano, 11 Dec 2000 16 DataGrid WP6 Meeting Piergiorgio Cerello Repeat with input/output file transfer
17
Milano, 11 Dec 2000 17 DataGrid WP6 Meeting Piergiorgio Cerello Test activities - Step2 Presently planned tests refer to the simulation stage ("Step1") What about reconstruction & analysis ("Step2")? More demanding from the point of view of GRID tools Need access to & shipping of more than 10 TB of distributed data
18
Milano, 11 Dec 2000 18 DataGrid WP6 Meeting Piergiorgio Cerello Conclusions Large number of sites participating to GRID tests & quoted for the PPR production (Step1) The use of GLOBUS as manager for the PPR production is becoming a reliable option Reconstruction & Analysis (Step2) for the ALICE Physics Performance Report will soon require more advanced GRID tools
19
Milano, 11 Dec 2000 19 DataGrid WP6 Meeting Piergiorgio Cerello
Similar presentations
© 2024 SlidePlayer.com. Inc.
All rights reserved.