1. The Region of Interest Strategy for the ATLAS Second Level Trigger
Véronique Boisvert
CERN
On behalf of the ATLAS Trigger/DAQ High Level Trigger Group
CHEP March 25th 2003

2. Outline Motivation Introduction The RegionSelector Tool Implementation
Timing Results
Conclusion

3. Motivation QCD Physics B Physics Higgs SUSY Top Exotics
+ ???  Discovery Potential!
SUSY
B Physics
QCD Physics
Higgs
Top
Exotics
Need fast, reliable, easy to make
Trigger Algorithms

4. Introduction bunch crossing rate: 40MHz 11m 23m
108 electronic channels
~1.6 MB/event
11m
23m

5. Introduction: Trigger Architecture

6. Introduction: Regions of Interest
RoI Mechanism:
Typically a few ROI / event
Only few % of event data required!
But more complex system

7. Sequence of Data Access
See: Architecture of ATLAS High Level Trigger Event Selection Software tomorrow
data store fetches requested collections only and caches them for subsequent requests
unit of retrieval from the DAQ is 1 ReadOutBuffer
possibility to do data preparation

8. Collections and Identifiers
See “Use of a generic identification scheme connecting events and detector description in the ATLAS experiment” on Thursday
Need to optimize collection granularity
Trade off between
judicious navigation for algorithms
Minimize data access requests
Use of offline identifiers
In process of validating design of using “offline” code into “online” environment
Facilitate development of algorithms
Study boundary between Level2 and EF
Allow performance studies for physics analyses
Needed for Inner Detector Regions
No Level 1 hardware identifier available

9. The Region Selector Tool
Algorithms at Level2 have access to Level1 RoI regions
Tool: Region in detector  Data collection identifiers
Requirements
Fast!
Different geometrical region descriptions
Cone
Region following a track
Etc.
Code to be run in multi-threaded environment

10. Implementation: The RegionSelector
At initialization
Fill 2 maps for each layer
Map < , set <identifiers>>
Map < identifier, vector<min, max>>
At Execution time: we want identifiers within < fmin, fmax> and <min, max>
Loop over the layers
Loop over f, get identifiers
Loop over the identifiers and check  interval
Return list of identifiers
No duplicate identifier ensured
EtaPhiMap

11. Implementation: data access granularity
Preliminary

12. Implementation: Geometrical Regions
CONE
Implemented for outer most detectors TRT, LAr, Tile and Muon
Simple: origin + h, Dh, f, Df
WEDGE OF CHEESE
Implemented for inner most detectors Pixel and SCT
Complicated: origin, Dz, h, Dh(r), f, Df
Realistic to account for LHC beam spread in z
HELICAL ROAD
Complicated: origin, 1/r (min), 1/r (max), tan q, f (min), f(max), h(min), h(max)
Trajectory-oriented (for uniform B field)
Future development

13. Timing Measurements Dh Improvements forthcoming Preliminary
Using 1 GHz Pentium 3
Numbers in ms
4 GHz scaled (div 3) DF

14. Timing Measurements In Process of timing algorithms as well
Offline way
Fully timing instrumented TestBeds
Running a Level2 LAr Calorimeter algorithm making use of Region Selector
Preliminary numbers: O(1 ms)
RegionSelector: ms
Small fraction of the algorithm time!
<< ~ 10 ms

15. Conclusion
The ATLAS High Level Trigger relies on the Region of Interest Mechanism to access only relevant part of the detector
The RegionSelector is the software tool used to extract the needed data collection identifiers based on some arbitrary geometrical regions
The timing measurements show a very minimal overhead on the algorithms coming from this tool
The RegionSelector tool will also be used for offline reconstruction of the data