1. The CMS High-Level Trigger
Leonard Apanasevich
University of Illinois at Chicago
On behalf of the CMS collaboration

2. Outline Introduction to the CMS Detector Trigger Challenges at the LHC
Trigger Architecture
Level-1 Trigger
High Level Trigger
HLT Processing Times
Trigger Commissioning with Cosmic Data
Trigger Rates and Menus for Startup
29 Sept, 2009
ICCMSE 2009

3. The Compact Muon Solenoid (CMS)
General purpose pp collider experiment:
Searches for Higgs bosons,
other new particles (SUSY,...) and new phenomena; Precision measurement of SM parameters like top and W masses, ...;
Heavy ion program.
For details see e.g.:
The CMS experiment at the CERN LHC, JINST 0803:S08004,2008;
29 Sept, 2009
ICCMSE 2009

4. The Large Hadron Collider
29 Sept, 2009
ICCMSE 2009

5. Trigger Challenges at the LHC
At design L = 1034 cm-2s-1
Interaction rate: ~1 GHz
~20 pp events per 25 ns crossing
1 kHz W events
10 Hz top events
< 104 detectable Higgs decays/year
Event size: ~1 MB (≈75M channels)
→ 1000 TB/sec for 1 GHz input rate
~300 MB/sec affordable
Enormous rate reduction necessary!
but without losing interesting physics
Select in stages:
Level-1 Triggers
1 GHz to 100 kHz (1/10000 reduction)
High Level Triggers
100 kHz to 100 Hz (1/1000 reduction)
29 Sept, 2009
ICCMSE 2009

6. Trigger Architecture
CMS has a two-tiered system to handle the LHC challenge
Level-1 trigger reduces rate from 40 MHz (xing freq) to 100 kHz (max)
High-Level triggers reduce rate from 100 kHz to O(100 Hz)
Front end pipelines
Readout buffers
Processor farms
Switching network
Detectors
Lvl-1
HLT
Lvl-2
Lvl-3
Front end pipelines
“Traditional”: 3 physical levels
CMS: 2 physical levels
Progress in networking/switching has justified CMS choice
29 Sept, 2009
ICCMSE 2009

7. Level 1: Only Calorimeter & Muon
Complex Algorithms
Huge amounts of data
Simple Algorithms
Small amounts of data
29 Sept, 2009
ICCMSE 2009

8. The Level-1 Trigger
Electrons, Photons, Jets, MET Muons
Reduce data rate from 40 MHz to 100 kHz while keeping the interesting physics events
Custom electronic boards and chips
Selects muons, electrons, photons, jets
ET and location in detector
Also Missing ET, Total ET, HT, and jet counts
Total decision latency: 3.2 s
128 Level-1 trigger bits. Bits can be set using logical combinations of Level-1 objects.
3<||< ||< ||< ||< <||<2.4 ||<1.2
29 Sept, 2009
ICCMSE 2009

9. Calorimeter Trigger Geometry
Trigger towers:  =  = 0.087
HCAL
ECAL HF
EB, EE, HB, HE map to 18 RCT crates
Provide e/g and jet, t, ET triggers
29 Sept, 2009
ICCMSE 2009

10. Muon Trigger Geometry DT: drift-tube system
CSC: cathode strip chamber system
RPC: resistive plate chamber system
Initial coverage of RPC is staged to <1.6
Initial coverage of CSC 1st station is staged to <2.1
29 Sept, 2009
ICCMSE 2009

11. Level-1 Trigger Menu
Determined by the physics priorities of the experiment
L1 Menu optimized to fit within the L1 bandwidth
Designed with a safety factor of 3 in mind
underestimation of input cross sections, poor beam conditions, detector performance, etc.
Realistic menu including double and mixed triggers for specific physics channels
Two L1 menus currently available
L = 8e29 cm-2s-1 (day-1 menu)
L = 1e31 cm-2s-1 (MC studies)
29 Sept, 2009
ICCMSE 2009

12. High Level Trigger Strategy
40 MHz
100 kHz
~150 Hz
29 Sept, 2009
ICCMSE 2009

13. High Level Trigger (HLT)
1 PB/s
High-Level triggers reduce rate from 100 kHz to O(150 Hz)
HLT does event reconstruction “on demand” seeded by the L1 objects found, using full detector resolution
Algorithms are essentially offline quality but optimized for fast performance
Farm of Processors
ONE event, ONE processor
High Latency
Simpler I/O
Sequential programming
300 MB/s
29 Sept, 2009
ICCMSE 2009

14. L2 unpacking (MUON/ECAL/HCAL)
HLT Algorithm Design
L1 seeds
L2 unpacking (MUON/ECAL/HCAL)
Local Reco (RecHit)
L2 Algorithm
Filter
L2.5 unpacking (Pixels)
L2.5 Algorithm
Each HLT trigger path is a sequence of modules
Processing of the trigger path stops once a module returns false
Reconstruction time is significantly improved by doing regional data-unpacking and local reconstruction across HLT
All algorithms regional
Seeded by previous levels (L1, L2, L2.5)
“Local”: using one sub-detector only
“Regional”: using small (η, φ) region
29 Sept, 2009
ICCMSE 2009

15. CMS Trigger/DAQ Evolution
CMS DAQ is a number of functionally identical, parallel, small DAQ systems
Build up 512512 switch from 8 6464 switches (DAQ slices)
HLT filter farm is currently comprised of 720 dual-quad core, 2.6 GHz, 16 GB memory Dell PE 1950 PC’s
7 out of 8 cores (~5000) dedicated to HLT processing
Remaining processor does event building, acts a resource broker, and other services.
29 Sept, 2009
ICCMSE 2009

16. m triggers: performance
Signal efficiency (eHLT/[eL1*A])
Background rates at L = 1032 cm-2 s-1
Threshold: single
m isolated
m non-isolated
Threshold: double
m non-isolated
Di-muon
Single muon
29 Sept, 2009
ICCMSE 2009

17. e/g triggers: performance
Background rates at
L = 1032 cm-2 s-1
Efficiency on benchmark channels (estimated from MC simulation)
Electrons:
Photons:
Electrons
Photons
29 Sept, 2009
ICCMSE 2009

18. Jets and Missing ET Jets: Rates for L = 1032 cm-2 s-1 ETmiss:
At HLT level, using an iterative cone algorithm with cone radius:
out of calorimeter “towers” (HCAL + projection on ECAL - deposits above certain thresholds)
ETmiss:
Algebraic sum of “transverse energies” (ET = E sinq) of calorimeter objects
Objects can be individual calorimeter cells
or jets
Can be used in combination with 1 or more jets
Rates for L = 1032 cm-2 s-1
29 Sept, 2009
ICCMSE 2009

19. t- and b-tagging at HLT Hadronic t decays: b-tagging triggers:
Level 1: seed from calorimeter towers in a narrow region
Level 2: HLT jet reconstructions and ECAL isolation
Level 3: Track isolation using pixel tracks in a smaller rectangular region or full tracks in a larger region
b-tagging triggers:
b-lifetime:
Jet requirements + global pixel track reconstruction to determine primary vertex at HLT level  requirement of at least 2 tracks with transverse impact parameter significance above a certain threshold
b → μ :
Jet requirements + L2/L3 muon reconstruction  requirements on DR(m-jet) and pT(m) relative to the jet direction
29 Sept, 2009
ICCMSE 2009

20. HLT Timing Considerations
HLT Timing is influenced by:
Trigger menu (L1T & HLT)
Determined by physics priorities
Input L1Trigger rate
Limited by bandwidth
Parameters of various L1T algorithms, e.g., H/E
HLT algorithms and configuration
Standard trigger paths at HLT seeded by L1 trigger bits
Order of modules and filters in a path
Parameters of the modules and filters
Allowable CPU performance:
L1 input rate / number of processors
@ 50 kHz: 100 ms/evt
@ 100 kHz: 50 ms/ evt
29 Sept, 2009
ICCMSE 2009

21. HLT Processing Times
Average time needed to run full Trigger Menu on L1-accepted events: 43 ms/event
Core Xeon processor running at 3.0 GHz
CPU times strongly dependent on HLT input
“Tails” have a significant impact on the average time
Will eliminate with time-out mechanism
29 Sept, 2009
ICCMSE 2009

22. Trigger Commissioning
CMS has collected over 300 million cosmic ray events
Can use these data for:
compare data and L1 emulation
check synchronization of signals from the various sub-detectors
Measure trigger rates from cosmic rays and from noise
Evaluate trigger efficiencies
Compare resolutions between trigger and reconstructed objects
29 Sept, 2009
ICCMSE 2009

23. Trigger Commissioning (II)
Synchronization of signals from different detector sub-systems demonstrated
L1 emulator: bit by bit software emulation of the L1 trigger
Excellent agreement between emulation and data
L1 trigger efficiency
as a function of offline reconstructed pT
29 Sept, 2009
ICCMSE 2009

24. Trigger Menu for Startup
Muons:
L1/L2/L3 muons run unprescaled at pT=20/9/3 GeV
Electrons:
No isolation requirements at L1
Unprescaled at pT=10 GeV with large pixel-matching window (LW)
Photons:
Unprescaled at pT=15 GeV with no isolation requirement
Jets:
No L1/HLT jet corrections applied at startup
Unprescaled at pT=30 GeV (~60 GeV corr.)
MinBias:
Several algorithms installed based upon HF-tower ET over threshold, HF ET ring sums, Ecal ET, and pixel triplets
Muon: 33 Hz
Electron: 30 Hz
Photon: 9 Hz
Jet: 36 Hz
MET & HT: 5 Hz
B-Tau: 4 Hz
MinBias: 14 Hz
Cosmics/Halo: 7 Hz
Total: 138 Hz
Trigger Rates
by Object
Rates for L=8e29 cm-2s-1
29 Sept, 2009
ICCMSE 2009

25. Summary LHC Trigger is Challenging
A new crossing every 25 ns with ~ 20 events at design luminosity  1 GHz of events input
All data stored 3 s then all but kHz rejected
Rejection of 99.99% of data without losing discovery physics!
Remaining event filtering done on a farm of CPUs running offline quality reconstruction algorithms
Rate of storage to archive is ~ 150 Hz
Rejection of % of data without losing discovery physics!!
Fast selection and high efficiency is obtained for the physics objects and processes of interest
Commissioning work already going on using Cosmic data
HLT and L1 have been developed for early luminosities
The overall CPU requirement is within the system capabilities
CMS trigger system is ready for collisions!!!
29 Sept, 2009
ICCMSE 2009