Presentation is loading. Please wait.

Presentation is loading. Please wait.

Future of Distributed Production in US Facilities Kaushik De Univ. of Texas at Arlington US ATLAS Distributed Facility Workshop, Santa Cruz November 13,

Similar presentations


Presentation on theme: "Future of Distributed Production in US Facilities Kaushik De Univ. of Texas at Arlington US ATLAS Distributed Facility Workshop, Santa Cruz November 13,"— Presentation transcript:

1 Future of Distributed Production in US Facilities Kaushik De Univ. of Texas at Arlington US ATLAS Distributed Facility Workshop, Santa Cruz November 13, 2012

2 Background  Distributed production requires many different ATLAS specific SW components/applications  Athena and Transformations – core software  ProdSys – task management system  AMI – Production Tags and Metadata  PanDA – job execution system  DQ2 – data management system  Monitoring of tasks, data and jobs  They utilize common tools like Globus, VDT, XRootD, Dcache, CVMFS, … deployed at our facilities Kaushik De 2November 13, 2012

3 Overview  Many distributed production components used in ATLAS are being upgraded after ~5 years of continuous use  In this talk we will focus on their evolution in 2013-2014  Athena on many fronts: AthenaMP, Athena64, AthenaGPU, AthenaPhi, Athena event service  trf -> tf  DQ2 -> Rucio  ProdSys -> ProdSys II  PanDA -> CAF  PanDA -> BigData  New monitoring capabilities Kaushik De 3November 13, 2012

4 AthenaXX  Many future paths for Athena driven by hardware – will not talk about them here  Interesting topic for distributed production – event service  Basic unit of measurement in HEP is events – not bits, bytes or files  Multi-core is the new paradigm (same as the old one)  Caching technologies may be best optimized at event level  Started discussions during SW week for event service  Client-server architecture in Athena desirable long term  PanDA server with Athena client will be first step to try November 13, 2012 Kaushik De 4

5 Job Transforms  Job transforms – trf – workflow wrapper around Athena  All production jobs use trf  Most major ATLAS workloads are supported  Including multi-step jobs  New workloads like overlay, FTK … are being added  Major changes underway  See recent talks by Graeme Stewart  https://indico.cern.ch/getFile.py/access?contribId=35&sessionId=19 &resId=0&materialId=slides&confId=169697 https://indico.cern.ch/getFile.py/access?contribId=35&sessionId=19 &resId=0&materialId=slides&confId=169697  https://indico.cern.ch/getFile.py/access?contribId=7&resId=0&materi alId=slides&confId=214562 https://indico.cern.ch/getFile.py/access?contribId=7&resId=0&materi alId=slides&confId=214562  Highlights of future changes in next few slides November 13, 2012 Kaushik De 5

6 November 13, 2012 Kaushik De 6

7 November 13, 2012 Kaushik De 7

8 November 13, 2012 Kaushik De 8

9 November 13, 2012 Kaushik De 9

10 November 13, 2012 Kaushik De 10

11 November 13, 2012 Kaushik De 11

12 November 13, 2012 Kaushik De 12

13 November 13, 2012 Kaushik De 13

14 November 13, 2012 Kaushik De 14 https://indico.cern.ch/getFile.py/access?contribId=1&sessionId=5 &resId=2&materialId=slides&confId=169697

15 November 13, 2012 Kaushik De 15

16 November 13, 2012 Kaushik De 16

17 November 13, 2012 Kaushik De 17

18 November 13, 2012 Kaushik De 18

19 What is ProdSys  Task management system  Interface to request production tasks  Generate jobs for execution by PanDA  Manage task completion  Consisting of many scripts  Web interface for task request  Bulk task submission interface  Auto generation of jobs from tasks  Scripts for task completion  Interacts with AMI and DQ2  And add-ons  Task-list creation scripts developed by production managers  Task monitoring November 13, 2012 Kaushik De 19

20 Current System November 13, 2012 Kaushik De 20 Production Manager Submits Tasks Jobs ProdSys Jobs PanDA User Bamboo User

21 What is ProdSys II  Split ProdSys into two parts  DEfT – task request and task definition  Some components will be taken from current ProdSys  JeDi – dynamic job definition and task execution  Integrated with PanDA (replaces Bamboo)  Will also be the engine for user analysis tasks  Need to work closely with Transforms & Rucio groups  All three systems should evolve together  Integration with monitoring  Will be planned from the beginning Kaushik De 21November 13, 2012

22 Future System November 13, 2012 Kaushik De 22 Production Manager DEfT PanDA User JeDi User

23 DEfT  Key features  Web UI for simplified interactive task request  Task request system based on physics requirements  Managers/users insulated from execution details  Deprecate/remove script based task submission  Error checking of task requests  Built-in authentication and approval mechanisms  Creates task according to a new simplified schema Kaushik De 23November 13, 2012

24 Tasks, Meta-tasks, Basket-tasks  New extensions to the concept of task  Task – basic unit  Input dataset -> Output dataset  Meta-task – chain of tasks, which will be auto-generated  Manager/user makes single request  Successive processing steps (transforms) created by DEfT  Intermediate steps in chain may be specified as transient  Basket-task – group of related tasks (eg. same tag)  Manager/user can define basket of tasks  Manager/user makes single request for execution  Ability to clone tasks, meta-tasks and basket-tasks  From pervious tasks, meta-tasks and basket-tasks  Or from predefined templates Kaushik De 24November 13, 2012

25 JeDi  Key features  JeDi will be core component of PanDA  Generate jobs dynamically from DEfT tasks  Jobs are defined to match execution environment and specified constraints(eg. number of cores, duration, file size, dataset size…)  Number of events varies per job  Jobs are not predefined with fixed number of events – key feature  PanDA responsible for optimal task execution  PanDA responsible for task completion  Auto-merging if requested  Data will be collected by PanDA to optimize job execution and completion (expanded concept of scout jobs) Kaushik De 25November 13, 2012

26 Common Analysis Framework  Task force to evaluate suitability of PanDA for a LHC common user analysis framework  Latest report: https://indico.cern.ch/getFile.py/access?contribId=7& sessionId=19&resId=1&materialId=slides&confId=16 9697 https://indico.cern.ch/getFile.py/access?contribId=7& sessionId=19&resId=1&materialId=slides&confId=16 9697 https://indico.cern.ch/getFile.py/access?contribId=7& sessionId=19&resId=1&materialId=slides&confId=16 9697 November 13, 2012 Kaushik De 26

27 November 13, 2012 Kaushik De 27

28 November 13, 2012 Kaushik De 28

29 November 13, 2012 Kaushik De 29

30 November 13, 2012 Kaushik De 30

31 November 13, 2012 Kaushik De 31

32 November 13, 2012 Kaushik De 32

33 Conclusion  Many updates/improvements planned 2013-2014  Some applications will be completely re-written  But based on past 5 years of LHC experience  Plans and teams are in place  Will lead to better software running at facilities  Waiting for current LHC run to end  Stay tuned for more November 13, 2012 Kaushik De 33


Download ppt "Future of Distributed Production in US Facilities Kaushik De Univ. of Texas at Arlington US ATLAS Distributed Facility Workshop, Santa Cruz November 13,"

Similar presentations


Ads by Google