High Level Trigger

1
High Level Trigger

• Trigger
• Accept/reject events
• Select
• Select regions of interest within an event
• Compress
• Reduce the amount of data required to encode the
  event as far as possible without loosing physics
  information
• Provide HLT-ESDs for online monitoring
• Access to the results of the event reconstruction

• Physics Applications
• Online
• Architecture
• Communication Framework
• Interfaces
• Prototypes
• Milestones

Dieter Roehrich UiB
2
Physics Applications

• Quarkonium spectroscopy
• Dielectrons
• Dimuons
• Open Charm
• Jets (C. Loizides, PhD thesis, Frankfurt, 2005)
• Pileup removal in pp

3
Quarkonium

• Dielectrons
• HLT task
• Reject fake TRD triggers and reduce trigger rate
  by factor of more than 10
• Status
• Fast TPC pattern recognition done
• Additional PID by dE/dx done
• Adaption of Kalman filter for HLT done
• Combined track fit TRD-TPC-ITS in progess
• To do
• Emulate the TRD Global Tracking Unit (TRD
  tracklet merging and PID)

T. Vik, PhD thesis, Oslo, 2005

• Dimuons
• HLT task
• Utilizing tracking chamber information and
  improving momentum resolution
• Sharpening of pt-cut
• Rejection factors low pt-cut 5, high pt-cut
  100
• Status
• Complete simulation including cluster finder
  done
• Full scale prototype HLT farm (UCT) done
• FPGA cluster finder in progress
• FPGA interface in progress

4
Open charm

• HLT task
• Detection of hadronic charm decays D0 ? K ?
• About 1 D0 per event (central Pb-Pb) in ALICE
  acceptance
• After cuts
• signal/event 0.001
• background/event 0.01
• Status
• Detailed study of timing profile of offline
  algorithm - done
• Adaption of ITS tracking to HLT and speed-up
  done
• Optimization of D0 finder in progress
• Combine HLT tracking and D0 algorithm in
  progress
• To do
• estimate the efficiency for appling
  D0-offline-cuts online
• extend study to D, D

5
Online

• Available modules
• TPC cluster finder (CF)
• TPC track follower (TF)
• Kalman fitter
• TPC Hough transform tracker (1)
• TPC Hough transform tracker (2)
• TPC cluster deconvolution
• TPC performance monitor
• TPC dE/dx (1)
• TPC dE/dx (2)
• TPC data compression (1)
• TPC data compression (2)
• ITS tracker
• Dimuon cluster finder
• Dimuon tracker
• Jet cone finder
• D0 finder

6
Tracking performance for CF/TF
Tracking efficiency Momentum resolution
Computing time 13 sec per event (dn/dy4000)
on a 1kSPECInt machine
A. Vestbø, PhD thesis, Bergen, 2004
7
Integral efficiency for CF/TF
Integral tracking efficiency
Contamination of fake tracks
8
Tracking performance for Hough transform
version 1

• Gray-scale Hough transform
• Image space raw ADC counts
• Transform space circle parameters
• Histogram increment charge

too CPU-time consuming
A. Vestbø, PhD thesis, Bergen, 2004
9
Tracking performance for Hough transform
version 2 (1)

• Linearized prehistoric Hough transform
• Image space conformal mapped cluster boundaries
• Transform space straight line parameters
• Histogram increment history of missing padrows,
  conditional

Collaboration with the Offline group Cvetan
Cheshkov
10
Tracking performance for Hough transform
version 2 (2)
Tracking efficiency
dN/dy8000 dN/dy6000 dN/dy4000 dN/dy2000 B0.5
T
Cvetan Cheshkov
11
Tracking performance for Hough transform
version 2 (3)

• Momentum resolution
• ?Pt/Pt(1.8xPt1.0) (B0.5T)
• ?(?)6.1mrad
• ?(?)5.5x10-3
• Computing time (1.3 kSpecInt machine)

Cvetan Cheshkov
12
ITS tracking (1)

• Offline tracking
• Modified offline code
• Speed-up of up to a factor of 30 for some modules

J. Belikov, C.Cheshkov
13
ITS tracking (2)

• Tracking efficiency

TPC only (HT) ITSTPC Fakes B0.5T
Comparable to offline
J. Belikov, C.Cheshkov
14
ITS tracking (3)

• Impact parameter resolution

Dominated by SPD -gt offline quality, i.e. 1
GeV/c track transverse impact parameter
resolution 60 microns
J. Belikov, C.Cheshkov
15
ITS tracking (4)

• Computing time (1.3 kSPECInt PC)

J. Belikov, C.Cheshkov
16
D0 finder

• Offline algorithm
• Cut on impact parameter
• calculate
• Distance of closest approach
• Invariant mass
• Decay angle
• Pointing angle
• Timing results (0.3 kSPECInt PC)

17
Data Compression - Principle
Standard loss(less) algorithms entropy encoders,
vector quantization ... - achieve
compression factor 2 (J. Berger et. al.,
Nucl. Instr. Meth. A489 (2002) 406)
Data model adapted to TPC tracking Store (small)
deviations from a model (A. Vestbø et. al., to
be publ. In Nucl. Instr. Meth. )
Cluster model depends on track parameters
Tracking efficiency before and after comp.
Relative pt-resolution before and after comp.
Tracking efficiency
Relative pt resolution
dNch/d?1000
18
Data Compression - Implementation

• Towards larger multiplicities
• cluster fitting and deconvolutionfitting of n
  two-dimensional response functions (e.g.
  Gauss-distributions)
• analyzing the remnant and keeping good clusters
• arithmetic coding of pad and time information

Leftovers
19
Data Compression - Results
Achieved compression ratios and corresponding
efficiencies
Compression factor 10
20
HLT Computer architecture

• HLT is a generic high performance cluster
• HLT-RORC
• HLT input/output
• FPGA co-processor

21
HLT-RORC design
22
FPGA co-processor

• Development framework
• VHDL testbench
• Hardware testbench
• Cluster finder
• VHDL design
• Fast linearized Hough transform
• Algorithm suited for FPGA implementation
• Design study has started

23
HLT CommunicationFramework

• Publish/Subscriber Frame-Work

24
HLT process control system

• Management of HLT processes
• Supervise processes
• React to state changes (e.g. errors, unexpected
  termination of processes)
• Management of system startup
• Orchestration of HLT system
• Requirements
• Flexible
• Hierarchical
• more than 2000 processes not manageable by single
  supervisor instance
• easy configuration
• Should be able to run on the cluster nodes
  themselves, along with the analysis processes
• Task Manager
• Architecture

25
HLT cluster control system
CHARM-Prototype
Name change CIA ? CHARM CHARM Cluster Hardware
Administration, Remote Management
Implemented features

• PCI scans
• Video card functionality
• VNC server (text mode)
• Web control
• Power control (computer)
• USB device mock-up
• Full memory access (PCI Bus)

CHARM went into preproduction W7/05
26
Interfaces

• HLT input
• HLT output
• DCS
• ECS
• Off-line

27
HLT data input / output

• Input HLT-RORC DIU firmware

• Output HLT-RORC SIU firmware

28
HLT DCS Monitoring interface

• HLT Communication Framework ported to DCS
  embedded system boards
• Access to / readout of TRD chips (TRAP) via slow
  control network
• Any data available on DCS can be inserted into
  HLT processing chain
• TCP Subscriber allows access from arbitrary
  programs, e.g. ROOT

HLT Node
HLT
DCS Embedded System
Pub/Sub TCP Bridge Receiver
HLT Component
Pub/Sub TCP Bridge Sender
DCS Hardware (e.g. TRAP chip or DCS sensors)
TCP Bridge Connection
DCS Readout Publisher
TCP Subscriber
PC/Workstation
ROOT
TCP Connection
29
HLT Prototypes

• TPC beam test (May/June 2005)
• TRD beam test (October 2005)
• Dimuon full scale prototype (November 2004)

30
TPC beam test

• Setup at T10

IROC
Detector LDC
HLT LDC
Detector LDC
DDL
Si BeamTelescope
D-RORC
D-RORC
VME processor CAEN VME boards
DDL
D-RORC
RCU 1
RCU 2
DDL
DDL
10 MB/s
DDL
Fast Ethernet
HLT
1 MB/s
Switch
Switch
HLT FEP
HLT FEP
HLT FEP
Gigabit Ethernet
Gigabit Ethernet
HLT
-
RORC
HLT
-
RORC
GDC
CASTOR 1.5 TB
3x 250 GB disk
HLT
-
OUT
HLT
-
OUT
HLT FEP
HLT FEP
HLT FEP
HLT FEP
HLT
-
RORC
HLT
-
RORC
HLT
-
RORC
Interfaces to DAQ successfully tested
31
TRD beam test
32
Dimuon full scale prototype
Capetown, November 2004 The real analysis
components (Dimuon tracker, Dimuon Decision Unit)
were simulated by dummy processes assuming real
CPU load and data sizes.
FilePublisher DiMuon Clusters
FilePublisher DiMuon Clusters
DiMuon Decision Unit
DummyLoad 0 DiMuon Decision Unit
DummyLoad 0 DiMuon Decision Unit
FilePublisher DiMuon Clusters
DiMuon Tracker
EventRate Subscriber (monitoring)
DummyLoad 0 DiMuon Tracker
DummyLoad 0 DiMuon Tracker
DummyLoad 0 DiMuon Tracker
DummyLoad 0 DiMuon Tracker
DummyLoad 0 DiMuon Tracker
FilePublisher DiMuon Clusters
1 process
3 processes on 3 processing nodes
6 processes on 6 processing nodes
FilePublisher DiMuon Clusters
15 nodes in total 1.2 kHz processing rate
10 processes on 5 processing nodes
Reading cluster data from file
Tracking
Trigger decision
33
Dimuon full scale prototype - Online GRID
Tromsø
Bergen
Dubna
Heidelberg
Successful deployment of on-line GRID in Nov.
2004, where detector data was processed in
real-time on-line.
Capetown
34
Milestones

• 542 RORC PPR (FPGA decision), Sep-05
• A number of suitable FPGAs were evaluated and a
  preliminary decision was taken.
• The H-RORC design is progressing and details are
  being optimized such as packaging and chip size
  versus cost. We expect to have a small number of
  prototype modules ready for the RORC PRR in
  September 2005.
• 543 HLT PRR (including network), Apr-06
• A number of HLT integration tests and data
  challenges have been and are currently being
  conducted. Most components are available to date.
  
• We expect to have a prototype available in due
  time for the PRR.
• 544 Start of HLT commissioning, Sep-06
• This milestone requires the first batch of all
  production modules to be available, allowing 4
  months for the production of the HLT specific
  hardware, such as the RORCs, which is feasible.
• 545 HLT with full connectivity - approx 30
  computing power, Apr-07
• The required compute power for the first running
  will be evaluated and decided by the
  collaboration. Noting that the HLT can be
  segmented into independent sub-parts, for
  instance serving a TPC sector each, the
  commissioning can be well done between Sept-06
  and April-07
• 546 HLT fully installed for HI (3 months prior
  scheduled full HI run), Apr-08
• This milestone basically requires the ordering
  and installing of the remaining computer
  resources (PCs) and their infrastructure, which
  does not present a particular challenge.