Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

The High Level Trigger and Express Data Production at STAR

Wayne Betts, Jinhui Chen, Yuri Fisyak, Hongwei Ke, Ivan Kisel, Pavel Kisel, Grigory Kozlov, Jeffery Landgraf, Jerome Lauret, Tonko Ljubicic, Yugang Ma, Spyridon Margetis, Hao Qiu, Diyu Shen, Qiye Shou, Xiangming Sun, Aihong Tang, Gene Van Buren, Iouri Vassiliev, Baoshan Xi, Zhenyu Ye, Zhengqiao Zhang, Maksym Zyzak

Abstract

To meet the demands of the Beam Energy Scan phase-II (BES-II) program, the STAR experiment at RHIC developed a dual real-time framework consisting of a High Level Trigger (HLT) and an Express Data Production system (xProduction). The HLT operates online within the Data Acquisition (DAQ) chain on a multicore CPU cluster, with optional acceleration using Xeon Phi coprocessors. It employs parallelized algorithms, such as the Cellular Automaton track finder, for fast tracking, vertexing, and event filtering, enabling real-time event selection and detector monitoring. In parallel, xProduction runs independently of the DAQ loop and performs near offline-quality calibration and reconstruction within hours. Using the express data stream, enhanced by HLT selections, and the STAR calibration framework, it enables early physics analysis and provides collaboration-wide access to analysis-ready datasets. Together, HLT and xProduction form a complementary system combining real-time selection with rapid high-quality reconstruction. This framework has enabled prompt reconstruction of the ${}^5_Λ\mathrm{He}$ hypernucleus and efficient processing of large datasets, demonstrating scalability for future high-luminosity experiments.

The High Level Trigger and Express Data Production at STAR

Abstract

To meet the demands of the Beam Energy Scan phase-II (BES-II) program, the STAR experiment at RHIC developed a dual real-time framework consisting of a High Level Trigger (HLT) and an Express Data Production system (xProduction). The HLT operates online within the Data Acquisition (DAQ) chain on a multicore CPU cluster, with optional acceleration using Xeon Phi coprocessors. It employs parallelized algorithms, such as the Cellular Automaton track finder, for fast tracking, vertexing, and event filtering, enabling real-time event selection and detector monitoring. In parallel, xProduction runs independently of the DAQ loop and performs near offline-quality calibration and reconstruction within hours. Using the express data stream, enhanced by HLT selections, and the STAR calibration framework, it enables early physics analysis and provides collaboration-wide access to analysis-ready datasets. Together, HLT and xProduction form a complementary system combining real-time selection with rapid high-quality reconstruction. This framework has enabled prompt reconstruction of the hypernucleus and efficient processing of large datasets, demonstrating scalability for future high-luminosity experiments.

Paper Structure

This paper contains 17 sections, 14 figures, 2 tables.

Table of Contents

  1. Introduction
  2. The High Level Trigger
  3. HLT Computer Farm
  4. Software Structure and Key Components
  5. Integration with STAR Data Acquisition
  6. Parallelization and Xeon Phi offload
  7. Real-Time Calibration
  8. Quality Assurance and Event Visualization
  9. Extended Functionality of the High Level Trigger (xHLT)
  10. Reconstruction Software
  11. Cellular Automaton Track Finder: Search for Long-Lived Particles
  12. Primary Vertex Search: Reconstruction of Collision Point
  13. KF Particle Finder: Search for Short-Lived Particles
  14. The Express Data Production
  15. BES-II: Search for Mesons and Hyperons
  16. ...and 2 more sections

Figures (14)

  • Figure 1: HLT event reconstruction pipeline. Information from sub-detectors, including TPC Anderson:2003ur, TOF Llope:2012zz, BEMC STAR:2002ymp, and Muon Telescope Detector (MTD) Yang:2014xta etc., are processed independently then jointly assembled to form a complete event to be used in trigger decision making.
  • Figure 2: Top: Scheme of HLT integration with Data Acquisition system at STAR. Bottom: Workload distribution in one HLT node. Because rates vary with beam energy and trigger mix, an illustrative benchmark is included : for Au+Au collisions at $\sqrt{s_{\rm NN}}$= 200 GeV the HLT sustained $\sim$ 2 kHz processing ($\sim$ 1.7-2 GB/s). BES-II input rates are lower; 200 GeV is shown as an upper-envelope stress case. Trigger-specific rates depend on the selection; e.g., the diMuon trigger had $\sim$ 1$\%$ acceptance.
  • Figure 3: The vertex positions in the x and y coordinates for events reconstructed by HLT (left) and offline (right) are shown. The ring structure is a result of collisions between the beam and the beam pipe.
  • Figure 4: Left: The energy loss per unit length as a function of primary momentum for tracks within the TPC, reconstructed by HLT. The primary momentum is the refit momentum with the primary vertex used as an additional hit. Lines are expected band centers for corresponding particle species; antiparticles are shown as mirrored bands. Right: The $1/\beta$ as a function of momentum for hits measured by TOF that are matched with a TPC track, reconstructed by HLT; particles and antiparticles are overlaid.
  • Figure 5: Architecture and data flow of HLT and xHLT. Left: DAQ integration. Detectors feed the event builders, which deliver events to HLT (L4); the L4 calibration server and online QA histograms run within HLT. HLT-good events are written to the distributed storage, while the full raw stream is logged to HPSS. Right: xHLT services on the HLT cluster. A MySQL database records job and file status; a Python production controller submits and tracks jobs for xCalibration, xProduction, and xPhysics; xProduction converts the buffered HLT-selected DAQ files to compact picoDsts, which are written on HLT storage
  • ...and 9 more figures