Download presentation
Presentation is loading. Please wait.
1
Grid Computing 6th FCPPL Workshop
Gang Chen & Eric Lançon March 29, 2013, NJU
2
Main achievements in Network performance monitoring and debugging Multi-core simulation tests on IHEP farm Workshop in Paris (June 2012) on site operational issues, 2 Chinese participants : SUN Gongxing/孙功星, YAN Xiaofei/闫晓 飞 Presentation of ZANG Dongsong/臧冬松 (PhD. in 2013) work at CHEP 2012 conference Student (LI Sha/李莎) spent 3 months in Europe to work on ATLAS data distribution system Visit to GRIF T2 (Paris) of YAN Xiaofei in December 2012 to discuss site configuration + Frequent meetings (ATLAS & LCG) with remote connection
3
Beijing T2 performance Thanks to high availability/reliability, Beijing T2 is classified ‘Direct’ T2 (T2D) Can get/send data to/from every T1/T2D site in the world Host primary data Network connection performance & stability are of primary concern T2D status may be lost if network deteriorates
4
Beijing site availability for ATLAS services
Well above 90% comp. element storage maintenance
5
Beijing site performance: data transferred
Import Beijing being T2D repository for ATLAS JET physics group 1PB (>1M files) Data volume transferred since March 2012 Export Beijing being T2D exports data to everywhere
6
Processing at Beijing over 90% job efficiency for centralized activities 50% of CPU consumption only for simulation ! Site is now also heavily used for user analysis, group analysis, reconstruction...
7
ATLAS Jobs through PanDA
Production Jobs: 311,5000 ( Job Success Rate: 94%) Analysis Jobs: 691,6000 (Job Success Rate: 81%)
8
Network performance monitoring
ATLAS ‘sonar’ : Calibrated file transfers by ATLAS Data Distribution system, from storage to storage perfSONAR (PS) : Network performance tool (throughput, latency), from memory to memory Has to be located as close as possible to storage at site and with similar hardware connectivity
9
perfSonar monitoring Deployment of perfSonar machines (Fazhi Qi/齐法制)
Work done in cooperation with GRIF T2 within the WLCG working group Identical configuration files for French and Chinese machines Monitoring hosted in BNL
10
ORIENT-plus : improved connectivity EU-China
11
Transfer rates from European T1s to Beijing
New Line : Not same impact for all T1s To be understood
12
Networking monitoring and debugging
Beijing connected to Europe via GEANT/ORIENT But performances not identical for all T1s, specific issues for each sites Asymmetries observed, not understood yet CERN → Beijing was using GLORIAD/KREONET changed end of 2012 on our request, ORIENT now used Firewall removed on our request at various sites Lyon→Beijing Beijing →Lyon KIT→Beijing Beijing →KIT
13
Multi-core processing
Special multi-core (8) queue setup at Beijing spring (YAN Xiaofei/闫晓飞), pioneers ! Used to validate AthenaMP, the ATLAS parallel event processing (to save memory) AthenaMP will be used as standard software for ATLAS simulation end of 2013
14
CMS Jobs Total jobs 829K : production 436k,analysis 231k
15
CMS Jobs production data : import 158TB,export 58TB
16
Prospect for Continue network monitoring and debugging activities Deployment of large multi-core setup for production, scaling issues to be addressed, common solutions with French T2s Deployment of WebDAV interface to storage (http access) in cooperation with French T2s Cloud computing : application for CSC-FCPPL grant for student (LI Sha/李莎) stay in Grenoble for 18 months Chinese-French-Japanese workshop at Beijing May 2013
17
THANK YOU Gang Chen/CC/IHEP 2018/12/5 - 17
Similar presentations
© 2025 SlidePlayer.com. Inc.
All rights reserved.