The High-Level Trigger of the ALICE Experiment

1
The High-Level Trigger of theALICE Experiment

• Heinz Tilsner
• Kirchhoff-Institut für Physik
• Universität Heidelberg
• International Europhysics Conference on
  High-Energy Physics 2003 Aachen

Further information http//www.ti.uni-hd.de/HLT
2
Content

• Physics Applications of the High-Level Trigger
• Online Pattern Recognition and Event
  Reconstruction
• Computing Infrastructure

3
Physics Applications
4
Physics Applications I

• Jet-Trigger
• Online jet trigger from TPC inspection of central
  Pb-Pb collisions at 200 Hz
• Cone jet-finder algorithm for online

• Pile-up removal
• Reconstruction of all tracks in the TPC
• Reconstruction of the event vertex
• Pile-up reduction by using a cut on impact
  parameter of tracks
• Data reduction about a factor 5

• Open-charm trigger
• Momentum filter (low pt cut)
• Examination of the event topology

5
Physics Applications II
Di-Muon Trigger Using information of the
di-muon spectrometers to determine the
transversal momentum ? pt cut
ee- Trigger
Reconstructing J/? and Y by their leptonic decays
into ee- pairs

• HLT reduces event rate about a factor of 10 by
• combining TRD tracklets with TPC and ITS
  tracking
• adding PID rejection power from TPC dE/dx

6
Online Event Reconstruction
7
HLT Functionality

• local pattern recognition (detector specific)
• cluster finder
• tracklets

• global pattern recognition
• e.g. global tracking in TPC

resulting trigger decision is based on fully
analyzed and reconstructed events
Time budget Online analysis needs 12s for one
eventwith dN/dY4000 or 2400 CPUs at event rate
of 200 Hz parallel processing on PC cluster
8
Fast Pattern Recognition

• low multiplicity events
• sequential feature extraction on space points
• cluster finder
• track follower

• high multiplicity events
• iterative feature extraction on raw data
• tracklet finder (Hough transform)
• parallel cluster evaluation

• FPGA co-processor
• releases CPU resources of host CPU
• online Hough Transform is essential for
  trackingin dense environment

9
Low Occupancy

• Cluster finder (FPGA)
• cluster finding
• centroid calculation
• deconvolution
• 2. Tracking (host CPU)

Hardware implementation
Verification of functionality C code VHDL
code

• Decoder
• decoding incoming ADC sequences (ALTRO list)
• calculating charge, sequence charge, and time
  ofa sequence
• Merger
• merges sequences of adjacent pads to clusters

10
High Occupancy Hough Transform
Hough Transform Transformation of coordinate
space (R, F) to parameter space (F0,
?) F0 emission angle? curvature
11
Hough Transform in FPGA Co-Processor
Behavioral (VHDL) model of Hough Transform
simulated and compared with software
12
Computing Infrastructure
13
FPGA Co-Processor as Part of the Front-End
Processor

• Front-End Prozessor
• First layer of the HLT-clusters
• Input for event data into the cluster (via
  optical link)
• normal PC, equipped with Read-Out Receiver
  Card (RORC)

• FPGA
• implementing the PCIbus protocol
• co-processor for online analysis

Ordinary PC cluster PCI RORC HLT
14
Data Volume Event Rates
15
HLT Cluster Setup
Example TPC sector
achieved event rate 430 events/s
16
Data Transport within the HLT-Cluster
(
Sub
) Event
Gatherer
new event
Subscriber
new event
Publisher
Subscriber
Subscriber
(Sub)Event Merger
Bridging between Nodes
Event m Block 0
Event m Block 1
Subs
Subs
Merging Code
Publisher
Event m Block 0, 1
17
Fault Tolerance

• Software framework with embedded fault tolerance
• Automatic re-configuration of the data path

A
B
Test setup with 7 computers
C
D
E

• Network connection disconnected
• Faulty PC node is removed from data path
• Spare node inserted into data path
• ? no single event is lost!

18
Prototypes

• 32 dual Pentium III PCsrunning Linux
• Network connection
• FastEthernet
• GigaBit Ethernet
• SCI

19
RORC Read Out Receiver Card
20
Summary

• HLT enables event selection based on physical
  signatures
• Online event analysis assisted by FPGA
  co-processor
• HLT allows for a significant reduction of the
  data volume
• Functional concept of the HLT exists
• Fault-tolerant software successfully tested

Further information http//www.ti.uni-hd.de/HLT