1 / 14

Offline Mass Data Processing using Online Computing Resources at HERA-B

Offline Mass Data Processing using Online Computing Resources at HERA-B. José Hernández DESY-Zeuthen. Motivation. Traditionally in HEP experiments, online and offline computing and software are sharply separated Different environment and requirements

hanley
Télécharger la présentation

Offline Mass Data Processing using Online Computing Resources at HERA-B

An Image/Link below is provided (as is) to download presentation Download Policy: Content on the Website is provided to you AS IS for your information and personal use and may not be sold / licensed / shared on other websites without getting consent from its author. Content is provided to you AS IS for your information and personal use only. Download presentation by click this link. While downloading, if for some reason you are not able to download a presentation, the publisher may have deleted the file from their server. During download, if you can't get a presentation, the file might be deleted by the publisher.

E N D

Presentation Transcript


  1. Offline Mass Data Processing using Online Computing Resourcesat HERA-B José Hernández DESY-Zeuthen J. Hernández. DESY-Zeuthen

  2. Motivation • Traditionally in HEP experiments, online and offline computing and software are sharply separated • Different environment and requirements • Dedicated hardware and software in DAQ and trigger • Commodity hardware (PC farms, Ethernet networks) and Linux OS, used in online environment, allow to blur the sharp border online/offline • HERA-B uses successfully Linux PC Farms in the Trigger and Data Acquisition systems • The reconstruction, typically offline task, is done online at HERA-B • HERA-B uses online computing and software resources to perform offline data reprocessing and MC production in the online PC Farms J. Hernández. DESY-Zeuthen

  3. HERA-B DAQ DSP switch High Bandwidth (10 Gbps) Low Latency (<10s) Online PC Farms J. Hernández. DESY-Zeuthen

  4. L2/L3 trigger step 240 nodes Intel Celeron 1.3 GHz 256 MB RAM Fast Ethernet NIC Linux OS No real time extensions CAN card for slow control Temperature, Power up/down Online reconstruction & L4 Trigger 100 dual-CPU nodes Intel PIII 550 MHz 256 MB RAM Fast Ethernet NIC Linux OS No real time extensions CAN card for slow control Temperature, Power up/down Online PC Farms L2/L3 Farm L4 Farm • Diskless PCs • PROM in NIC loads Linux • Extremely ease maintenance • DSP-to-PCI interface • data link to DSP switch (40 MB/s, 1s driver latency) J. Hernández. DESY-Zeuthen

  5. L4 FARM tasks • Full online event reconstruction • Allow immediate physics analysis • Avoid relatively slow access to tape (20 TB/year) • Full online reconstruction allows online Data Quality Monitoring and Online Calibration and Alignment • Online Event Classification and Selection • Mark events in physics categories (event directories) • L4 trigger step • Data logging • Add reconstruction info to event and send to logger J. Hernández. DESY-Zeuthen

  6. L4 FARM Software • Linux environment • Process server • Frame Program ARTE • Reconstruction, analysis and MC • Same code online and offline • Data I/O shm memory based (online) and file based (offline) • Event reconstruction time ~ 4 sec  50 Hz output rate Process Server J. Hernández. DESY-Zeuthen

  7. Online DQM and CnA • Online CnA to keep trigger performance & online reconstruction • DQM from reconstructed data • Gathering system to increase statistics • CnA version tag in event data • CnA constants multicasted to L2 nodes by DAQ • CnA constants retrieved from DB by L4 nodes when new CnA tag in events J. Hernández. DESY-Zeuthen

  8. Booting and State Machine • Each run has ~2000 process. (~ 400 are under State Machine). • The run is booted in 3 minutes (~10 process/s ). • Different machine types: Linux, Lynx and DSP. & the same protocol. • The State Machine maps different transition tables @ different levels in the State Machine tree. • All procresses are booted remotely in different machines using the messaging system. J. Hernández. DESY-Zeuthen

  9. Offline  Online • Idea: Use online idle time to perform offline mass data processing using the online computing resources • Shutdown periods, time between spills, accelerator down time • Use vast online computing resources • 440 CPUs, high network bandwidth • Use not only online hardware but also online processes and protocols: • Use online boot and control systems • Use online data transmission protocols • Perform “online” Data Quality Monitoring • Run “quasi-online” Data re-processing and Monte Carlo production J. Hernández. DESY-Zeuthen

  10. Data Taking L2 Buffers EVC DSP switch L2/L3 Farm TAPE L4C Ethernet switch Archiver L4 Farm J. Hernández. DESY-Zeuthen

  11. Data Re-processing L2/L3 Farm TAPE Ethernet switch Archiver L4 Farm Provider J. Hernández. DESY-Zeuthen

  12. Monte Carlo Production • Full Monte Carlo Production: • Generation, Detector Simulation, Digitization, Trigger Simulation and Full Reconstruction • 30 sec/evt @ 1.3 GHz node. 300 KB/evt  1 Million evts/day, 300 Gbytes/day L2/L3 Farm TAPE Ethernet switch Archiver L4 Farm J. Hernández. DESY-Zeuthen

  13. Quasi-Online Processing • System fully integrated in the Run Control System • Shift crew can use efficiently the online idle time • Same online processes and protocols used for booting, control, monitoring, data reconstruction, data quality, logging and archiving Data Reprocessing = Online Reconstruction J. Hernández. DESY-Zeuthen

  14. Summary • Efficient use of online computing resources at HERA-B to perform mass offline data processing • Not only the online hardware is used but also the online boot, control, monitoring and data transmission processes and protocols LHC experiments might consider to include the online computing power as GRID resources in order to use the online idle time for offline mass data processing J. Hernández. DESY-Zeuthen

More Related