Presentation is loading. Please wait.

Presentation is loading. Please wait.

ATLAS – statements of interest (1) A degree of hierarchy between the different computing facilities, with distinct roles at each level –Event filter Online.

Similar presentations


Presentation on theme: "ATLAS – statements of interest (1) A degree of hierarchy between the different computing facilities, with distinct roles at each level –Event filter Online."— Presentation transcript:

1 ATLAS – statements of interest (1) A degree of hierarchy between the different computing facilities, with distinct roles at each level –Event filter Online processing No reprocessing during shutdown periods, but maybe in the future –Tier-0 Prompt reconstruction of calibration and express streams. RAW data archive –Tier-1 (about 10 centers – collaboration wide services) Partial RAW data archive (low latency) Provide access to processed data Reprocessing and backup for Tier-0 primary event processing Scheduled analysis Pledged resources and reliable operation for the whole collaboration (24/7) –Tier-2 (about 30 centers) Monte Carlo simulation Analysis facility Calibrations

2 ATLAS – statements of interest (2) Data processing with the ATLAS software framework Storage technology processed data: POOL ROOT files –RAW event size 1.6MB, output rate 200Hz –Processed event size 500kB/100kB, Tag data 1kB/event –Maximum file size 2GB (RAW) / 625MB (processed) Each of the facilities is responsible for funding the resources to fulfill its role ATLAS will negotiate relationships between Tier-1 and Tier-2 centers in order to optimize the system in terms of data transfer, storage and network Sliding scale, Tier-1 and Tier-2 centers of different sizes. However, resource ratio of a center type is fixed Significant contribution of Tier-3 centers not included

3 ATLAS – statements of interest (3) Tier-2 specific remarks Each ATLAS user has resource quota on these centers. –Only visible resources are credited by ATLAS accounting. Provide analysis capacity for working groups –This analysis activity is chaotic in nature (not scheduled) Center hosts the tag data and part of the processed data –Shared Tier-2 replica of Tier-1 processed data Each event is written to exactly 1 stream, but may be referenced from more than 1 tag collection –i.e. somebody else might want these events Provide simulation capacity, migrate data to Tier-1 Replication models and policies may be negotiated among Tier-1 and associated Tier-2 centers

4 ATLAS – the numbers Assumed identical ramp-up for Tier-1 and Tier-2 centers, but this might be delayed for Tier-2 centers (following luminosity) Projections do not include resource replacements After 2010 about 30% resource increase per year Sum of all Tier-1 centers2007200820092010 CPU (MSI2k|nodes 2007)7.9 | 260026.5 | 880047.6 | 1590081.3 | 27000 Storage tape|disk (PB)3.0 | 5.510.1 | 15.518.5 | 23.130.9 | 41.9 Network in | out (MB/s)-100 | 19 140 | 23 Cost/(MCHF | M€)18.7 | 12.523.1 | 15.414.8 | 9.921.0 | 14 Sum of all Tier-2 centers2007200820092010 CPU (MSI2k|nodes 2007)7.3 | 240021.1 | 700031.9 | 1060052.2 | 18000 Storage tape|disk (PB)0 | 3.20 | 10.10 | 17.00 | 26.6 Network (MB/s)-10++ Cost (MCHF | M€)11.0 | 7.314.5 | 9.78.5 | 5.78.7 | 5.8

5 LHCb – statements of interest (1) A degree of hierarchy between the different computing facilities, with distinct roles at each level –Online farm (1800 CPU’s) Online processing Reprocessing during shutdown periods –Tier-0 Distribution of data in quasi real-time RAW data archive –Tier-1 (about 6 centers + CERN – collaboration wide services) Partial RAW data archive – except for CERN Primary event processing Provide access to processed data Event reprocessing Distributed analysis Pledged resources and reliable operation for the whole collaboration (24/7) –Tier-2 (about 14 centers) Monte Carlo simulation Not foreseen as analysis facility, but not proscribed either

6 LHCb – statements of interest (2) The majority of the distributed analysis will be performed at CERN and the Tier-1 centers Sliding scale, some Tier-2 centers may not qualify as signatories for the MoU. Storage technology processed data: POOL ROOT files –RAW event size 25 kB, output rate 2kHz –Processed event size 75kB/25kB Reprocessing once per year, during shutdown period –The online farm takes care of halve of the reprocessing –The transfer rate to/from the pit is the highest during the reprocessing period Stripping twice per year, parallel with experiment Monte Carlo production on Tier-2 centers is an ongoing activity throughout the year. –Data is migrated to the Tier-1 centers

7 LHCb – the numbers For Tier-1 centers, the numbers after the “/” are for the reprocessing during the shutdown period. The numbers between ( ) are during the stripping months For Tier-2 centers, the numbers between ( ) are applicable for analysis centers Financial numbers are scaled from Atlas (do not include replacement) Sum of all Tier-1 centers2007200820092010 CPU (MSI2k|nodes 2007)2.7 | 9004.4/7.4 (8.4) | (2800) 5.6/8.5 (10.8) | (3600) 8.4/8.5 (?) | 2800 Storage tape|disk (PB)1.2 | 1.52.1 | 2.44.3 | 2.97.1 | 3.4 Network in | out (MB/s)-17.7 (91) | 11.0 (76) 17.7 (89) | 11.0 (60) 31.0 (?) | 22.0 (?) Cost (MCHF | M€)5.7 |2.2 |1.6 |2.6 | Sum of all Tier-2 centers2007200820092010 CPU (MSI2k|nodes 2007)4.6 | 15007.7 (8.0)| (2700) Storage tape|disk (TB)0 | 14.00|23.0 (200) Network in | out (MB/s)-0.4 (50) | 1.1 ? Cost (MCHF | M€)2.5 | 1.71.1 | 0.70 | 0

8 ALICE – statements of interest (1) Although ALICE starts with a hierarchical model, it is less pronounced than for the other two experiments. Eventually it should evolve into a cloud model, where service levels and functionality are the only distinctive features between centers –Tier-0 Primary event processing RAW data archive Calibration and Alignment Fast access to reconstructed data –Tier-1 (6 Tier-1 centers + CERN) Partial RAW data archive – except for CERN Reconstruction, event processing and scheduled analysis Fast access to reconstructed the reconstructed data of this center Fast access analysis and simulated data of this center and its Tier-2 centers –Tier-2 (21 Tier-2 centers) Monte Carlo simulation Data reduction and user analysis Fast access to simulated data and analysis data of this center

9 ALICE – statements of interest (2) Data processing with AliEn (Alice Environment) –Based on web services –Aim: transparent user access to worldwide computing resources –Tight integration with ROOT (Parallel ROOT Facility – PROOF) Pb – Pb interactions are not streamed to the Tier-1 centers in quasi real-time. This is not viable, since data is recorded at 1.25 GB/s –Event size 2 GB (p-p events only 20 MB) Not much difference in data streams between first year and subsequent years due to installation of selective triggers User analysis on a Tier-2 center usually involves temporarily hosting the data required for this analysis 2 reprocessing passes per year Tier-1 provides reliable storage for its Tier-2 centers

10 ALICE – the numbers Numbers for Tier-1 centers based on 1 month Pb – Pb run + 3 month transfer The numbers for the Tier-2 centers are fixed by the requirements to process all the Monte Carlo data and to perform the end-user analysis tasks Financial numbers are scaled from Atlas (do not include replacement) Sum of all Tier-1 centers2007200820092010 CPU (MSI2k|nodes 2007)2.8 | 9005.5 | 180013.8 | 460016.6 | 5500 Storage tape|disk (PB)1.5 | 1.53 | 37.5 | 7.57.7 | 8.5 Network in | out (MB/s)-2000 | 20 ? Cost (MCHF | M€)6.0 | 4.03.6 | 2.47.6 | 5.11.1 | 0.7 Sum of all Tier-2 centers20072008Begin 20092010 CPU (MSI2k|nodes 2007)2.7 | 9005.5 | 180013.7 | 460016.4 | 5500 Storage tape|disk (PB)0 | 0.50 | 10 | 2.60 | 2.9 Network in | out (MB/s)-10 | 600 ? Cost (MCHF | M€)2.6 | 1.71.7 | 1.13.3 | 2.20.6 | 0.4

11 Summary Networks are not included All experiments expect about 3 Tier-2 centers per Tier-1 1 ATLAS Tier-2 center2007200820092010 CPU (nodes 2007)80250350600 disk (TB)100300600900 Cost (k€)250300200 1 LHCb Tier-2 center2007200820092010 CPU (nodes 2007)110200 disk (TB)115 Cost (k€)120500 | 0 Note on long term replacement: All experiments indicate that they can live with a replacement of 30% of the total capital investment per year. Suggestion: multiply by 3 to install NIKHEF integrated Tier-2 center 1 ALICE Tier-2 center20072008Begin 20092010 CPU (nodes 2007)50100220300 disk (TB)2550125150 Cost (k€)805010020 Total cost (k€)14501200900660


Download ppt "ATLAS – statements of interest (1) A degree of hierarchy between the different computing facilities, with distinct roles at each level –Event filter Online."

Similar presentations


Ads by Google