Upload
jed
View
19
Download
1
Embed Size (px)
DESCRIPTION
The Athena Control Framework in Production, New Developments and Lessons Learned. September 27 2004 C. Leggett , P. Calafiura, W. Lavrijsen, M. Marino, D. Quarrie. Athena and Gaudi. Gaudi framework shared by LHCb, ATLAS, GLAST, HARP, and OPERA - PowerPoint PPT Presentation
Citation preview
NATIONAL ENERGY RESEARCH SCIENTIFIC COMPUTING CENTER
Charles Leggett <[email protected]>
The Athena Control Framework in Production, The Athena Control Framework in Production, New Developments and Lessons LearnedNew Developments and Lessons Learned
September 27 2004
C. Leggett, P. Calafiura, W. Lavrijsen, M. Marino, D. Quarrie
2Charles Leggett <[email protected]>
9/2
7/2
00
4
Athena and GaudiAthena and Gaudi Gaudi framework shared by LHCb, ATLAS,
GLAST, HARP, and OPERA Based on a modular component architecture,
structured around dynamically loadable libraries Maintains a separation between transient and
persistent layers, allowing new technologies to replace aging ones with minimal impact on the end user
Athena comprises the ATLAS specific extensions to Gaudi, most notably:
Storegate – the data store Interval Of Validity Svc – managing time dependent
data Pileup – combining multiple events HistorySvc – maintaining a multi-level record Python scripting
3Charles Leggett <[email protected]>
9/2
7/2
00
4
Athena and GaudiAthena and Gaudi
Converter
Algorithm
StoreGateSvc
PersistencyService
DataFiles
Algorithm
StoreGateSvc PersistencyService
DataFiles
Detector Store
MessageService
JobOptionsService
Particle Prop.Service
OtherServices
HistogramService
ApplicationManager
ConverterConverterEventLoopMgr
AuditorsScriptingService
Sequencer
Event Store
DDDDD
DDDD
HHH
HH T TT
4Charles Leggett <[email protected]>
9/2
7/2
00
4
Athena in ProductionAthena in Production
Data Challenge II: phase 1:
~30 Physics channels, 10s of millions of events several million calibration events currently producing raw data in a distributed worldwide environment using Grid
phase 2: reconstruction and real time distribution of data to tier 1 institutes
phase 3: worldwide distributed analysis on the Grid Combined TestBeam
taking data since July in various configurations ~5000 runs, > 1Tb data written G4 simulation and reconstruction of CTB setup
occurring in parallel conditions databases in production, both read and
write. preparing for phase II: massive reconstruction of all
real data and production of MC data.
6Charles Leggett <[email protected]>
9/2
7/2
00
4
Interval of Validity ServiceInterval of Validity Service
Makes associations between user data and time dependent data that resides in specialized conditions databases
Transparent to user Data only read from persistent layer when it is
used. Validity interval information is separate from data
Hierarchical callback functions can be associated with time dependent data such that they are triggered when data enters a new interval of validity
Validity interval information and time dependent data can be preloaded on a job or run basis for trigger or testbeam situations where database access is unwanted
7Charles Leggett <[email protected]>
9/2
7/2
00
4
Access to Time Varying DataAccess to Time Varying Data
Maintains separation of transient and persistent layers
Testbeam environment making good use of IOVService for:
Slow control Calibration
8Charles Leggett <[email protected]>
9/2
7/2
00
4
Detector Pileup in DC2Detector Pileup in DC2
Overlay ~1000 min bias events to original physics stream
Requirement: digitization algorithms should run unchanged
Tuple event iterator: manage multiple input streams
Select random permutations from a circular buffer of min-bias events
Memory optimization: requirement total job size < 1GB2-dim detector and time-dependent event caching
Stress test architecture flexibility
Excellent tool to expose memory leaks (they become x1000 bigger)
9Charles Leggett <[email protected]>
9/2
7/2
00
4
HistoryHistory
Provenance of data must be assured
User selection of data based on its history. full history of generation and processing recorded and
associated with all data
Important in analysis to know complete source of data, and all cuts applied
History Service keeps track of Environment Job configuration Services Algorithms, AlgTools, SubAlgorithms DataObjects
10Charles Leggett <[email protected]>
9/2
7/2
00
4
Python Based Scripting InterfacePython Based Scripting Interface Python woven into the framework, replacing flat
text configuration files
dynamic job configuration conditional branching detFlags
interactive analysis
data object access and manipulation
connection to ROOT histogramming facilities
object type info for dictionaries and persistency
11Charles Leggett <[email protected]>
9/2
7/2
00
4
Detector Configuration MatrixDetector Configuration Matrix
-+- 28 A DetFlags.Print()
: pixel SCT TRT em HEC FCal Tile MDT CSC TGC RPC Truth LVL1
detdescr : ON ON ON -- -- -- ON -- -- -- -- ON --
digitize : ON ON ON -- -- -- ON -- -- -- -- ON --
geometry : ON ON ON -- -- -- ON -- -- -- -- ON --
haveRIO : -- -- -- -- -- -- -- -- -- -- -- -- --
makeRIO : -- -- -- -- -- -- -- -- -- -- -- -- --
pileup : ON ON ON -- -- -- -- -- -- -- -- ON --
readRDOBS : -- -- -- -- -- -- -- -- -- -- -- -- --
readRDOPool : ON ON ON -- -- -- ON -- -- -- -- ON --
readRIOBS : -- -- -- -- -- -- -- -- -- -- -- -- --
readRIOPool : -- -- -- -- -- -- -- -- -- -- -- -- --
simulate : -- -- -- -- -- -- -- -- -- -- -- -- --
writeBS : -- -- -- -- -- -- -- -- -- -- -- -- --
writeRDOPool : ON ON -- -- -- -- ON -- -- -- -- ON --
writeRIOPool : -- -- -- -- -- -- -- -- -- -- -- -- --
12Charles Leggett <[email protected]>
9/2
7/2
00
4
Lessons LearnedLessons Learned
Design for Performance pileup excellent testbed database access can be problematic
Design for Persistency various container classes rewritten with persistency in
mind ClassID Service to globally monitor objects
Better support for realtime monitoring essential for proper testbeam studies