Скачать презентацию ALICE Activities in Service Challenge 3 P Cerello Скачать презентацию ALICE Activities in Service Challenge 3 P Cerello

1563242e17dff262390f3ca8d94576ff.ppt

  • Количество слайдов: 15

ALICE Activities in Service Challenge 3 P. Cerello (INFN – Torino) Nov 28 th, ALICE Activities in Service Challenge 3 P. Cerello (INFN – Torino) Nov 28 th, 2005 1 January 18, 2004 Piergio Cerello ([email protected] infn. it)

ALICE 2005 Physics Data Challenge o Number of events (preliminary, they will increase) n ALICE 2005 Physics Data Challenge o Number of events (preliminary, they will increase) n Simulation o o o n - 24, 000 central) - 60, 000 central) - 1, 000 central) 85, 000 central Reconstruction o n 30, 000 Pb-Pb (equivalent to: 100, 000 p-p (equivalent to: 6 -12 h x -> 0. 5 - 1 Mh Much quicker -> 15 Kh Assume 1, 000 CPUs: o o Reconstruction: 1 day Simulation: 0. 5 -1 Kh each (25 -50 days occupancy) November, 28 th, 2005 T 2 Meeting - Torino 2

ALICE 2005 Physics Data Challenge o Physics Data Challenge n Coordinate with SC 3 ALICE 2005 Physics Data Challenge o Physics Data Challenge n Coordinate with SC 3 to run the ALICE Physics Data Challenge in the SC 3 framework November, 28 th, 2005 T 2 Meeting - Torino 3

ALICE & LCG Service Challenge 3 o Primary Goals: n n n Use of ALICE & LCG Service Challenge 3 o Primary Goals: n n n Use of the deployed LCG SC 3 infrastructure for the ALICE Data Challenge 2005 Test of data transfer and storage services (SC 3) Test of distributed reconstruction and calibration model (ALICE) Integrate the use of LCG resources with other resources available to ALICE within one single VO interfaced to different “grids” Analysis of reconstructed data (SC 4) November, 28 th, 2005 T 2 Meeting - Torino 4

ALICE & LCG Service Challenge 4 o Primary Goals: n Interactive Analysis of reconstructed ALICE & LCG Service Challenge 4 o Primary Goals: n Interactive Analysis of reconstructed data with PROOF November, 28 th, 2005 T 2 Meeting - Torino 5

ALICE & LCG Service Challenge 3 o Use case 1: RECONSTRUCTION n n n ALICE & LCG Service Challenge 3 o Use case 1: RECONSTRUCTION n n n (Get “RAW” events stored at T 0 from our Catalogue) First Reconstruct pass at T 0 Ship from T 0 to T 1’s (goal: 500 MB/S out of T 0) Reconstruct at T 1 with calibration data Store/Catalogue the output November, 28 th, 2005 T 2 Meeting - Torino 6

ALICE & LCG Service Challenge 3 o Use Case 2: SIMULATION n n Simulate ALICE & LCG Service Challenge 3 o Use Case 2: SIMULATION n n Simulate events at T 2’s Transfer Data to supporting T 1’s November, 28 th, 2005 T 2 Meeting - Torino 7

ALICE and LCG Service Challenges o In other words: n n n Mimic our ALICE and LCG Service Challenges o In other words: n n n Mimic our data (raw-like + simulated) flow test the reconstruction and the prototype calibration framework measure the performance of the SC 3 services/components: o o n n n Data transfer efficiency Storage Element efficiency Start in September 2005 As new sites keep coming in, increase the scale of the exercise As new middleware comes in, test it & add more functionality November, 28 th, 2005 T 2 Meeting - Torino 8

ALICE & LCG Service Challenge 3 o What do we need for SC 3? ALICE & LCG Service Challenge 3 o What do we need for SC 3? n n n n n VO-box deployment on SC 3 sites - LCG/ALICE ROOT/Ali. Root deployment on SC 3 sites - ALICE Ali. En Top Level Services - ALICE WMS (n RBs) + CE/SE Services on SC 3 - LCG LFC instances on all SC 3 sites, seen from WNs/VO-boxes - LCG glite. FTS accessible from all WNs/VO-boxes - LCG SRM (DPM) accessible from all WNs/VO-boxes - LCG SC 3 resources for ALICE (Computing/Storage) - LCG Appropriate JDL files for the different tasks - ALICE November, 28 th, 2005 T 2 Meeting - Torino 23

SC 3: Status of deployment ü ü Ali. En Central Services Deployment on SC SC 3: Status of deployment ü ü Ali. En Central Services Deployment on SC 3 sites ü ü n n n T 1 s: CERN, CNAF, CC-IN 2 P 3, Grid. Ka T 2 s: Bari, Catania, GSI, ITEP, JINR, Torino Ongoing: SARA, RAL (issue on allocated storage) Ali. ROOT deployment: automated (Ali. En Pack. Man) UI(s) for submission to LCG/SC 3: on each VO-box WMS (n RBs) + CE/SE Services on SC 3 LFC instances on all SC 3 sites, seen from WNs/VO-boxes glite. FTS accessible from all WNs/VO-boxes SRM (DPM) accessible from all WNs/VO-boxes November, 28 th, 2005 T 2 Meeting - Torino 24

Results o Distribution of running jobs in November 2005 by site November, 28 th, Results o Distribution of running jobs in November 2005 by site November, 28 th, 2005 T 2 Meeting - Torino 25

Results o Distribution of completed jobs in November 2005 by site November, 28 th, Results o Distribution of completed jobs in November 2005 by site November, 28 th, 2005 T 2 Meeting - Torino 26

Results o Sample of network outgoing traffic by site November, 28 th, 2005 T Results o Sample of network outgoing traffic by site November, 28 th, 2005 T 2 Meeting - Torino 27

ALICE & LCG Service Challenge 3 o o ALICE PDC 2005 is also aimed ALICE & LCG Service Challenge 3 o o ALICE PDC 2005 is also aimed at evaluating the performance of SC 3 services in realistic conditions Partial results are quite satisfactory in terms of stability and performance (i. e. , much better than in 2004) thanks to n n o o Improved LCG stability Agent-based approach Until the end of SC 3, it will be a stability/scalability exercise Focus on full functionality rather than just data transfer n Minimised Data Transfer is “integrated” in the ALICE data model November, 28 th, 2005 T 2 Meeting - Torino 28

ALICE & LCG Service Challenge 4 o o Focus on Distributed (Interactive) Data Analysis ALICE & LCG Service Challenge 4 o o Focus on Distributed (Interactive) Data Analysis Issues: n Query efficiency to the Metadata Catalogue o n Site stability is mandatory o o n Now it is Central, will it become distributed? The system must be fault tolerant (I. e. analyse a subsample of the requested input and normalise properly) Site services must be available, ideally, 24 x 7 System reaction to a large number of users is unknown, however our model should: o Minimise the impact on data transfer November, 28 th, 2005 T 2 Meeting - Torino 29