1. STAR Level-3 C. Struck CHEP 98 1 Level-3 Trigger for the Experiment at RHIC J. Berger 1, M. Demello 5, M.J. LeVine 2, V. Lindenstruth 3, A. Ljubicic, Jr. 2, D. Roehrich 1, E. Schaefer 6, J.J. Schambach 4, D. Schmischke 1, M.W. Schulz 2, R. Stock 1, C. Struck 1,a, P. Yepes 5 (1) University of Frankfurt (2) Brookhaven National Lab., Upton, NY (3) University of Heidelberg (4) University of Texas at Austin (5) Rice University, Houston, TX (6) Max-Plank-Institut fuer Physik, Munich (a) Yale University, New Haven, CT Christof Struck August, 98 University of Frankfurt Yale University

2. STAR Level-3 C. Struck CHEP 98 2 Solenoidal Tracker At RHIC Au+Au at  s = 200 GeV/nucleon pair 5000-10000 charged particles/event polarized p+p at  s = 500GeV baseline detector: large TPC STAR Experiment

3. STAR Level-3 C. Struck CHEP 98 3 STAR DAQ TPC organized in 24 geometrical sectors; each sector delivers digitized data through 6 readout boards (custom built VME boards); same readout scheme for SVT (4 sectors) and FTPC (6 sectors) readout boards equipped with: –buffer for 12 uncompressed events –processing power (three i960 per board) STAR DAQ hierarchical system of VME systems - interconnected by a low latency and high bandwidth network: SCI: event building, inter-crate communication/synchronization and for Level-3 data- passing Global VME Crate SCI Ring TPC Sector CrateSVT Sector Crate 24 crates4 crates SCSI

4. STAR Level-3 C. Struck CHEP 98 4 Level-3 Requirements Au+Au collisions: TPC event rate after Level-0 Trigger: 100 Hz, expected event size: 100 MByte  2.0 GByte/sec after zero suppression event size 20 MByte  20 MByte/sec 10  8 bit translation zero suppression Level-3 Tape / Offline

5. STAR Level-3 C. Struck CHEP 98 5 Level-3 Trigger software trigger: select events according to –event topology, kinematics –specific signature selecting sub-events: store only raw data of interesting tracks (regions-of-interest, ROI), e.g. tracks of lepton candidates data compression process the raw data: perform pattern recognition in real time at 100 Hz Reduce data by a factor of 100, therefore use knowledge of physics of these collisions. Possibilities:

6. STAR Level-3 C. Struck CHEP 98 6 Software Trigger or Regions-of-Interest Estimated rates for J/   e + e – and   e + e –. 100 Au+Au collisions/sec L3-Trigger sensitivity 1:100 assumed S eff = S / (1+2B / S)

7. STAR Level-3 C. Struck CHEP 98 7 Online Pattern Recognition Reconstruction of full event including track merging between different detectors max. TPC event rate 100 Hz  average time for one event  10 msec DAQ receiver boards provide buffer for 12 uncompressed raw events  Level-3 processing time  120 msec tracking quality not as high as offline analysis, but precise enough to enable fast trigger decision Input: raw data of slow detectors: TPC, SVT, FTPC raw data of fast detectors: CTB, MWC, VTC, VPD, ZDC, EMC

8. STAR Level-3 C. Struck CHEP 98 8 Concept of Level-3 scalable hierarchical structure of processors pattern recognition done sequentially DAQ readout boards: cluster finding transform raw ADC data into space coordinates Sector Level-3: track finding perform track finding on sector level Global Level-3: collect all information from local nodes, merge tracks and make trigger decision

9. STAR Level-3 C. Struck CHEP 98 9 Cluster Finder I runs on DAQ readout board processors (Intel i960) input: beginning and ending time of pixel sequences in each pad prepared in receiver board ASICs output: space coordinates and charge of found clusters optimized for speed includes deconvolution of merged clusters

10. STAR Level-3 C. Struck CHEP 98 10 Cluster Finder II Simulation: Au+Au, Venus + GEANT

11. STAR Level-3 C. Struck CHEP 98 11 Cluster Finder Results Simulation: Au+Au, Venus + GEANT pad row 16 - 20

12. STAR Level-3 C. Struck CHEP 98 12 Cluster Finder Timing

13. STAR Level-3 C. Struck CHEP 98 13 Track Finder combines a number of space points to form track segments track segments are merged to form vertex and non-vertex tracks provides: –particle momenta –particle identification via dE  dx algorithm : (P. Yepes at CHEP ‘97) –conformal mapping to speed up fitting procedures –optimized data organization and memory management results: –efficiency between 80 - 90 % for p T > 0.4 GeV/c and pseudorapidity  < 1.2; comparable to offline track finder –momentum resolution  p T / p T 0.4 GeV/c (vertex constraint)

14. STAR Level-3 C. Struck CHEP 98 14 Track Finder Results

15. STAR Level-3 C. Struck CHEP 98 15 Track Finder Timing timing for one TPC sector simulated central Au+Au, Venus

16. STAR Level-3 C. Struck CHEP 98 16 Architecture distributed, symmetric, scalable processor system OS and Software: WinNT (or Linux/ Solaris) for processing nodes; VxWorks for DAQ nodes; software is written in C/C++ SCI network connection between local sector units and global system to provide high bandwidth (poster, J. Schambach, CHEP ‘98) baseline configuration : 12 local processing clusters later: 24 TPC sectors, 4 SVT sectors and 6 FTPC sectors

17. STAR Level-3 C. Struck CHEP 98 17 Architecture: Sector Level-3 SB 1 SB n      DAQ SCI Ring SL3 1/1 Level-3 local SCI Ring 1 SL3 machine 1 processor Alpha 21264 SL3 1/4 SL3 SMP machine 4 processors Quad Pentium II Level-3 local SCI Ring n GL3

18. STAR Level-3 C. Struck CHEP 98 18 Architecture: Global Level-3 GL3 Broker SB 1 SB n      EVB DAQ SCI Ring GL3 1/4 Level-3 global SCI Ring GL3 SMP machine 1 4 processors GL3 n/4 GL3 SMP machine n 4 processors Token Manager To Trigger System

19. STAR Level-3 C. Struck CHEP 98 19 Examples for Level-3 Au+Au collisions Select events with J/  candidates in e + e – -channel –find electron candidates –loop over electron pair candidates »calculate mass with vertex constraint »select events in mass window (e.g. 2.5 - 4 GeV) p+p collisions –remove pile-up in TPC: select trigger event out of 600 - 800 visible events in the TPC –select events based on threshold for jets, photons and electrons

20. STAR Level-3 C. Struck CHEP 98 20 Data Compression I zero suppression general data compression methods, loss-free (e.g. Huffman encoding) or lossy reduce only by a factor of 2 to 5 later phase of the experiment data modeling techniques reduce by a factor up to 15, keep only relevant information: –assume data model for cluster and tracks (helix for STAR) –store only quantized differences to data model of found tracks –pattern recognition can be redundant –detector performance has to be well understood !!

21. STAR Level-3 C. Struck CHEP 98 21 Data Compression II track parameter for a helix model cluster parameter

22. STAR Level-3 C. Struck CHEP 98 22 Data Compression III typical event: 8000 tracks with 45 cluster each (Venus simulation, worst case)  track data: 8000 32 Byte = 0.24 MByte  cluster data: 8000 45 3 Byte = 1.03 MByte  total: 1.3 MByte needed (lower limit) compare to -raw data (zero suppressed) : 20 MByte -cluster data: 2.7 MByte (8 Byte per cluster)

23. STAR Level-3 C. Struck CHEP 98 23 Summary / Outlook Level-3 Trigger is needed for most of the STAR physics programs reduction of data rate by a factor of 100 requires pattern recognition in real time at 100 Hz processed event rate of about 15 - 20 Hz can be achieved using the shown fast algorithms and either one Alpha 21264 or a Quad Pentium II (sufficient for year one) system is scalable by adding more CPUs  higher event rates What needs to be done? choose processors and OS –therefore timing results for Alpha 21264 needed prototype setup full simulation of trigger scenarios using the Level-3 chain (cluster finder + tracking)