Presentation is loading. Please wait.

Presentation is loading. Please wait.

Karol Buńkowski CERN, UW Trigger at CMS Warsaw 2 November 2011.

Similar presentations


Presentation on theme: "Karol Buńkowski CERN, UW Trigger at CMS Warsaw 2 November 2011."— Presentation transcript:

1 Karol Buńkowski CERN, UW Trigger at CMS Warsaw 2 November 2011

2 2 Overview Trigger architecture –L1 Trigger –High Level Trigger Muon trigger performance Karol Buńkowski, CERN, UW, CMS L1 Trigger Review, 11 October 2011

3 Christian HartlThe CMS Level-1 Trigger @ SPG-ÖPG 2011, Lausanne3 CMS at LHC

4 P. Klabbers, U. Wisconsin, TIPP, 9 June 2011 CMS Level-1 Trigger - 4 CMS Detector * *Actually 3.8 T

5 5 http://press.web.cern.ch/press/PressReleases/Releases2011/PR22.11E.html Karol Buńkowski, CERN, UW, Warsaw, 2 November 2011

6 6 Luminosity at CMS in 2011 Karol Buńkowski, CERN, UW, Warsaw, 2 November 2011 2011: √s = 7 TeV Peak Instantaneous Luminosity: 3.69 ×10 33 cm -2 s -1 1380 bunches/beam (1331 colliding at CMS) minimum collisions spacing - 50 ns Primary vertices – up to ~20 (mean ~10)

7 7 Why trigger is crucial at LHC? Warszawa, 5 października 2009 Karol Buńkowski, UW p-p collisions 40 millions of bunch crossing/s: ~20 p-p collisions each crossing and hundreds of secondary particles ~1 MB of compressed data/BX = 4  10 13 Bytes (40 000 GB) / s Not possible to record! The background - Standard Model physic - must be rejected in the real time (without recording to the massive storage) The signal – Higgs or “new physic” particles - must be kept with efficiency close to 100% New particles are heavy  decays to the energetic objects which allow to distinguish them from the background events The rate of the selected events cannot exceed the data acquisition bandwidth

8 8 Trigger system - principles The Trigger decides in the real time whether each event should be recorded to the massive storage (because is potentially interesting) or should be rejected (because most probably is background) The Trigger is the first step of the physic analysis of the detector data: –Cuts (filters) which should allow to distinguish the signal (“new physic”) from the background (Standard Model) are applied. But opposite to the offline analysis the events (class of events) rejected by the Trigger are definitely lost. If the events of “new physic” are not accepted by the Trigger we might never see them… We don’t know exactly what we are looking for, therefore the Trigger must be universal: should accept the possibly wide variety of events (but keeping the rate within the limit) Karol Buńkowski, CERN, UW, Warsaw, 2 November 2011

9 The Trigger and DAta Aqusition system at CMS Readout buffers 128 events = 3.2  s Level 1 Trigger Dedicated electronics (ASICs, FPGAs) @ 40 MHz, only logic functions Analyses every event (bunch crossing, BX)  pipeline processing; latency 3.2  s, including ~2  s for data transmission between the detector and counting room, dead time free operation Output ≤ 100 kHz High Level Trigger (HLT) Computer Farm: 1008 nodes, 9216 cores, 16 TB memory runs the software events selection algorithms A few hundreds of Hz recorded on the magnetic tapes Event Builder - switching network. Gathers the data from one event into one HLT computer Karol Buńkowski, UWWarszawa, 5 października 2009 Coarse data Detector keep reject DAQ: readouts the data for the selected events, the events are fragmented

10 Level 1 Trigger architecture 4  4+4  MIP+ ISO bits L1A (trigger) 40 MHz pipeline ECAL Trigger Primitives ECAL Trigger Primitives HCAL Trigger Primitives HCAL Trigger Primitives Regional Calorimeter Trigger Regional Calorimeter Trigger RPC hits CSC hits DT hits Segment finder Track finder Track finder Pattern Comparator Pattern Comparator Segment finder Track finder Track finder Global Trigger TTC system TTS system Detectors Frontend Status Link system 32 partitions Karol Buńkowski, UW Muon TriggerCalorimeter Trigger Trigger subsystems: identify, measure and sort the trigger objects e/ , J, E T, H T, E T miss Warszawa, 5 października 2009 muons Global Calorimeter Trigger Global Calorimeter Trigger Global Muon Trigger

11 P. Klabbers, U. Wisconsin, TIPP, 9 June 2011 CMS Level-1 Trigger - 11 L1 Trigger Custom Hardware Hundreds of boards Thousands of: ASICs FPGAs Copper Cables Optical Fibers (Wo)man hours Global Muon Trigger & Global Trigger (Vienna) RCT (Wisc.) CSCTF (Florida) DTTF (Vienna) GCT (Imperial) RPC PaC (Warsaw)

12 12 L1 Global Trigger conditions and algorithms GT input - Trigger Objects: From Global Muon Trigger –4 best muons (with p T, position, sign, and quality information) From Global Calorimeter Trigger - energy and coordinates of: –4 forward and 4 central jets, –4  -jets, –4 isolated and 4 non-isolated e/ , –total E T, missing E T, –H T (the scalar sum of the transverse energies of the jets above a threshold ), –threshold-dependent jet multiplicities 64 technical trigger signals (from LHC beam counters, CMS beam scintillators, CMS subsystems) The GT first calculates the “conditions” : For each quadruplet of “particlelike” (muons and jets) channels checks if: – p T or E T above threshold –quality above threshold –η or φ within window –absolute difference in η or/and φ between two particles is within a required range Cuts on the total E T, missing E T and H T are applied Several conditions are then combined by a simple combinatorial logic (AND-OR-NOT) to form algorithms (128 are possible) The algorithms are combined by a final OR function to generate a ‘L1_Accept’ - signal that starts the Data Acquisition Karol Buńkowski, CERN, UW, Warsaw, 2 November 2011

13 13 L1 Trigger “menu” – example algorithms Muon triggers: –L1_SingleMu14_Eta2p1 p T >= 14 GeV |  | < 2.1 GMT quality > 3 (low quality unconfirmed CSC candidates are rejected) –L1_DoubleMu3: p T >= 3 GeV Electron/gamma triggers: –L1_SingleEG15: E T >=15 Gev –L1_DoubleEG_12_5: DoubleIsoEG_12GeV_5GeV OR DoubleNoIsoEG_12GeV_5GeV OR (SingleIsoEG_12GeV AND SingleNoIsoEG_5GeV) OR (SingleIsoEG_5GeV AND SingleNoIsoEG_12GeV) Jet triggers –L1_DoubleJet44_Central: DoubleCenJet_44GeV OR DoubleTauJet_44GeV OR (SingleCenJet_44GeV AND SingleTauJet_44GeV) Cross-triggers –L1_Mu12_EG5 SingleMu_12GeV AND ( SingleIsoEG_5GeV OR SingleNoIsoEG_5GeV ) Usually new menu = new GT firmware (the thresholds are programmable registers, but conditions must be rebuilt in the firmware) Karol Buńkowski, CERN, UW, Warsaw, 2 November 2011

14 14 HLT system CMS online filter farm (hardware) –1008 nodes (system boards – computers): 2 x 4 cores or 2 x 6 cores (2.66 GHz), 16 or 24 GB RAM –9216 cores in total Each machine runs The HLT software: –1 master process input/output watchdog re-spawns event processors –7 event processes - HLT reconstruction and selection: read the input data run all the trigger algorithms – “paths” (443 in the latest proton menu) take the final accept/reject decision stream the data to the Storage Managers For processing one event each process has available an average up to 90 ms (For comparison, offline reconstruction takes ~5 s per event!) Karol Buńkowski, CERN, UW, Warsaw, 2 November 2011

15 15 HLT paths and menu The HLT decision is taken as the OR of many independent triggers or “paths” (>400). Each path runs independently from the others (in parallel in the same process - the software guarantees that the same reconstruction block is not run twice) Most of the paths requires one selected Level 1 algorithm (as a seed) to start Then step-by-step reconstruction is performed Reconstruction is seeded by the Level 1 object –First stage: only calorimeters and muon system information (fast code) –Second stage: reconstruction of full tracks in the tracker (slowest code) –Intermediate stage: use of partial tracker information try to reject the event as soon as possible The even is taken, if at least one path passed Karol Buńkowski, CERN, UW, Warsaw, 2 November 2011

16 16 Karol Buńkowski, CERN, UW, Warsaw, 2 November 2011

17 17 Data streams and datasets The HLT decision is used to steer splitting events into streams and datasets online: –Stream A: bulk data (sample for physics analysis) –Alignment & Calibration (AlCa) streams –express (prompt feedback & calibrations) offline: split the streams into 20 Primary Datasets: Jet, HT, MET, BTag, MultiJet, SingleMu, DoubleMu, MuOnia, MuEG, MuHad, SingleElectron, DoubleElectron, Photon, ElectronHad, PhotonHad, Tau, TauPlusX, Commissioning, Cosmics, MinBias Karol Buńkowski, CERN, UW, Warsaw, 2 November 2011

18 18 Prescaled triggers It is good idea to save some fraction of events which are not accepted by the standard “physic” cuts: no cuts at all, except requirements for any collisions – minimum bias events events triggered with lower thresholds These triggers are pre-scaled, i.e. only every n event with given trigger path fired is accepted. n is prescale factor. These triggers are utilized for: fake rate measurements efficiency measurements trigger rate monitoring trigger debugging/development Collecting special data for detector calibration and performance evaluation Main types: L1 pass-through's L2 pass-through's L3 low p T paths Physics triggers with looser ID criteria Physics triggers with looser isolation or without isolation As the luminosity was increasing, the trigger thresholds were increased as well. The lower threshold triggers were kept, however they were prescaled Karol Buńkowski, CERN, UW, Warsaw, 2 November 2011

19 19 Dynamic prescales The beam intensity decreases during a fill  the luminosity (number of interactions) decreases  the rate of the events accepted by the Trigger decreases  the available bandwidth of the data acquisition is not used. Therefore the prescales (both at L1 and HLT) are decreased to keep approximately constant trigger rate. Karol Buńkowski, CERN, UW, Warsaw, 2 November 2011

20 20 How to cope with prescales? If in a given analysis the events taken with the prescaled triggers are used, the prescales must be included in calculation of the cross-section (or rate) of the process under consideration: –The simplest approach is to take events triggered by only one trigger (HLT) path and weight each event with the prescale factor that was applied in given lumisection *. The prescales factors for each lumisection (as well as the luminosity) are available in the CMS database –If more than one HLT paths needs to be taken, the overlap of the path must be included * lumisection = 23.14 s Karol Buńkowski, CERN, UW, Warsaw, 2 November 2011

21 21 Trigger efficiency measurement (watch out for bias…) How to find the denominator (it must include the events NOT triggered by given path)? In case of the L1 efficiency the additional problem is the filtering applied by the HLT – bias Possible approaches: –Take the events in which the required object(s) were reconstructed from the minimum bias dataset – the simplest approach, but very low statistic, especially for the more energetic objects –Take the events triggered by other paths not correlated with the investigated path, e.g. for calculation of the muon trigger efficiency take the events triggered by jet triggers, check if the muon trigger fired for the events with the reconstructed muons (might be biased), –Data driven methods (check for which efficiency the Monte Carlo fits best to the real data), –Use tag-and-probe method Karol Buńkowski, CERN, UW, Warsaw, 2 November 2011 # events with trigger path fired Trigger path efficiency = # all events with the object(s) required by the trigger

22 22 Tag and probe method Use the Z or j/psi mass resonance to select lepton (electron or muon) pairs: Tag: lepton passing very tight selection with very low fake rate (<<1%) Probe: lepton passing softer selection and pairing with Tag object in a way that the invariant mass of tag and probe combination is consistent with the Z (j/psi) resonance Efficiency = N pass /N all N pass → number of probes passing the selection criteria (e.g. muon or electron trigger fired) N all → total number of probes counted using the resonance Karol Buńkowski, CERN, UW, Warsaw, 2 November 2011

23 23 Muon trigger performance Karol Buńkowski, CERN, UW, Warsaw, 2 November 2011

24 24 Muon trigger objects Karol Buńkowski, CERN, UW, Warsaw, 2 November 2011 L1 Muon Trigger 3 muon detectors to |eta|<2.4 Drift Tubes (barrel) and Cathode Strip Chambers (endcaps) track segment identification Track Finder Resistive Plate Chambers Pattern Matching 4 candidates per subsystem to Global Muon Trigger mergers, removes duplicates and sorts candidates, 4 top candidates to Global Trigger HLT Muon Trigger

25 25 Muon triggers thresholds ad rates the latest proton menu Lowest p T unprescaled triggers, rates at luminosity ~3e33 cm -2 s -1 L1 –L1_SingleMu14_eta2p1, rate ~6 kHz –L1_DoubleMu3p5, rate ~5 kHz HLT –HLT_Mu40 (single), rate ~8.5 Hz –HLT_IsoMu24 (single), rate ~13 Hz –HLT_Mu13_Mu8 (double), rate ~5 Hz –HLT_DoubleMu5_IsoMu5, rate ~0.1 Hz Very high cut on the single muon. High selection efficiency of (for example) W bosons require a muon threshold around 20 GeV, and therefore cannot be achieved with the single muon trigger alone. Double muon and various cross triggers are required. Double muon trigger efficiency is about 90% Karol Buńkowski, CERN, UW, Warsaw, 2 November 2011

26 26 L1 Muon trigger efficiency Karol Buńkowski, CERN, UW, Warsaw, 2 November 2011 CSC Track Finder efficiency Tag and probe, Z mass, |eta| < 2.4

27 27 High Level Trigger Efficiency for Muon 30 GeV Path HLT_Mu30 No isolation requirements on trigger X-axis is offline muon P T Efficiency is measured using the tag and probe technique with events selected offline with two muons forming a Z ➞ μμ candidate Inefficiencies result mainly from L1 muon requirements –HLT efficiency in plateau is 95% in barrel and 90% in endcap –HLT / L1 efficiency is 99% in both MC and data Barrel Endcaps

28 28 Trigger on HSCPs on L1 Karol Buńkowski, CERN, UW, Warsaw, 2 November 2011 Heavy Stable Charged Particles (HSCPs) take as much as 2 BX to exit the detector –Look for late signal in muon triggers –But preBPTX veto (the beam pick-up signals – technical trigger) during 50 ns bunch spacing vetoes these! RPC Trigger can extend detector signal to 2 BX –Reduce RPC delay into GMT by 1 BX –preBPTX will veto early in-time muons –In-time muons still captured with collision BX –HSCPs aligned with collision BX –Create special HLT path to keep slow particles (the BXid will be +1 wrt to Tracker, for example). HSCPs Triggering! layer 6 layer 5 layer 4 Chamber hitsextended hits BX Muon candidate In-time muon layer 3 layer 2 layer 1 BPTX layer 6 layer 5 layer 4 Chamber hits Extended hits BX Muon candidate HSCP layer 3 layer 2 layer 1 BPTX

29 29 Bibliography Andrea Bocci (CERN), The CMS High Level Trigger: operations and data taking experience, Parallel given at CHEP2010: International Conference on Computing in High Energy and Nuclear Physics 2010, 18-22 Oct 2010, Taipei (Taiwan) Malgorzata Kazana (Soltan Inst. for Nucl. Studies), CMS trigger and data taking in 2010, Invited given at Epiphany 2011: Cracow Epiphany Conference on the First Year of the LHC, 10-12 Jan 2011, Institute of Nuclear Physics PAN, Kraków (Poland), Joao Varela, Triggers for physics at instantaneous luminosity 1E33 in the CMS experiment, Invited given at RDMS CMS Conference: 15th Annual RDMS CMS Collaboration Conference, 22-28 May 2011, Kharkov, Pamela Renee Klabbers (Univ. of Wisconsin), Operation and Performance of the CMS Level-1 Trigger during 7 TeV Collisions, Contributed given at TIPP 2011: Technology and Instrumentation in Particle Physics 2011, 9-14 Jun 2011, Chicago, IL (United States), Christian Hartl (CERN), The CMS Level-1 Trigger, Contributed given at SPS-ÖPG11: Joint Annual Meeting of the Swiss Physical Society and the Austrian Physical Society, 15-17 Jun 2011, EPFL, Lausanne, Vaud (Switzerland), Karol Buńkowski, CERN, UW, Warsaw, 2 November 2011

30 30 Backup Karol Buńkowski, CERN, UW, Warsaw, 2 November 2011

31 P. Klabbers, U. Wisconsin, TIPP, 9 June 2011 CMS Level-1 Trigger - 31 e/  and Jet Algorithms 4x4 Tower sums from RCT to GCT Jet or  E T 12x12 trig. tower  E T sliding in 4x4 steps w/central 4x4 E T > others  : isolated narrow energy deposits Energy spread outside  veto pattern sets veto Jet   if all 9 4x4 region  vetoes off GCT uses tower sums for E T,ME T jets for H T, MH T Electron (Hit Tower + Max) 2-tower  E T + Hit tower H/E Hit tower 2x5-crystal strips >90% of E T in 5x5 (Fine Grain) Isolated Electron (3x3 Tower) Quiet neighbors: all towers pass Fine Grain & H/E One “L” of 5 EM E T < Thr.

32 P. Klabbers, U. Wisconsin, TIPP, 9 June 2011 CMS Level-1 Trigger - 32

33 33 Karol Buńkowski, CERN, UW, Warsaw, 2 November 2011 Integrated muon rates at generator level from different sources. High luminosity L = 10 34 cm - 2 s -1. Limited to |  | < 2.1

34 34 For any considered “new physic” channel the Trigger configuration (L1 and HLT) must be found which: –Will be sufficiently efficient in selecting the interesting events, –The rate will not be grater than bandwidth given to you, –Is possible to implement in the L1 (only simple cuts and logic functions are possible) and HLT (the time needed to perform the algorithm is within the limit). Karol Buńkowski, CERN, UW, Warsaw, 2 November 2011


Download ppt "Karol Buńkowski CERN, UW Trigger at CMS Warsaw 2 November 2011."

Similar presentations


Ads by Google