150 likes | 302 Views
Online Reconstruction. Linda R. Coney 5 November 2009. Online Reconstruction. G4MICE uses the unpacker to look at data from DATE It then converts the raw data into information with physical meaning Goal: Provide a fixed set of histograms to be filled during data taking
E N D
Online Reconstruction Linda R. Coney 5 November 2009 Linda R. Coney – 5 November 2009
Online Reconstruction • G4MICE uses the unpacker to look at data from DATE • It then converts the raw data into information with physical meaning • Goal: • Provide a fixed set of histograms to be filled during data taking • These histograms will contain quantities that can give information about the physics happening – a first look at analysis quantities • Provides another data quality check • Are we taking the data we think we are? • Are the detectors & beam behaving as planned? • Provide graphical interface to display plots • Provide comparison plots for shifters • Not meant to be final results • Collaboration chooses list of useful histograms • Preliminary list follows Linda R. Coney – 5 Nov 2009
Online Reconstruction Histograms • Step I • TOF • # planes hit/event • # slabs hit/event • Pulse heights • Distribution in x, y across TOF0, TOF1, TOF2 • 2D x vs y g gives shape of beam • Reconstructed time-of-flight • Momentum if assume p etc • dt (plane 1 vs plane2) in TOF0, TOF1, TOF2 • 2D x vs y g gives shape of beam • RF phase calculated vs absolute time in spill • CKOV • Light yield in CKOVA and in CKOVB • Construct some CKOV measure using the two CKOVs • KL ??? • PID determination • CKOV A B light yield vs TOF • TOF vs PTracker • Check x,y in TOF1 vs x,y in the tracker at input and similarly at the other end of the channel for TOF2 and tracker2 Linda R. Coney – 5 Nov 2009
Online Reconstruction Histograms • Step II & Beyond • Tracker(s) • # planes hit/event • # stations hit • Pulse heights • # points used in online reconstruction • Muon px, py, pz, pT, p at the 2 tracker reference planes • x,x’, y,y’ • 1D, 2D plots of position at 2 tracker reference planes • Light yield distributions for each station • PID determination • CKOV A B light yield vs TOF • TOF vs PTracker • Check x,y in TOF1 vs x,y in Tracker1 at input and same for TOF2 and exit of Tracker2 • EMR ??? Linda R. Coney – 5 Nov 2009
Online Reconstruction Histograms • Step I & Beyond • Beam emittance, amplitude • Beamline 2D plots (x, x’) & (y,y’) • x,y • x’,x • y’,y • ex, ey • momentum • MICE Step III and IV – energy in Tracker1 vs energy in Tracker2 and DE • MICE Step V and VI • DE vs RF phase • DE vs absolute time Linda R. Coney – 5 Nov 2009
Online Reconstruction Histograms • What is needed to produce these plots? • Online Reconstruction farm • G4MICE installed on farm • TOF reconstruction • CKOV reconstruction • Tracker reconstruction • KL reconstruction • EMR reconstruction • Calibration for all detectors • Unpacking code for each detector • Check that G4MICE uses unpacker in a same way that Online Monitoring uses unpacker Linda R. Coney – 5 Nov 2009
Current Status of Reconstruction • TOF Reconstruction and calibration well underway • CKOV reco same • Tracker reconstruction works Linda R. Coney – 5 Nov 2009
Online Reconstruction Farm • Installed two farm computers in MICE control room March 09 • MICEOnRec01a • MICEOnRec01b • MICEOnRec02a • MICEOnRec02b • Total of three quad-core processors • G4MICE installed on both • Tests run • Reconstructed tracker cosmic ray test data • 114 events/second • Ran simulation, digitization, and reconstruction of Step VI • Simulation: ~262 events/second • Simulation + Digi: ~236 events/second • Reconstruction: ~1920 events/second Linda R. Coney – 5 Nov 2009
Online Reconstruction Histograms • What is needed to produce these plots? • Online Reconstruction farm • G4MICE installed on farm • TOF reconstruction • CKOV reco • Tracker reco • KL reco ? • EMR reco (later) • Unpacking code for each detector • TOF, CKOV, GVA, KL • Trackers, EMR (late 2009) • Check that G4MICE uses unpacker in same way that Online Monitoring uses unpacker • Can produce online monitoring plots with G4MICE • Testing under way to compare to standard Online Monitoring plots (11/09) Linda R. Coney – 5 Nov 2009
Online Reconstruction • Read data • Questions: Read from socket (as Online Monitoring) or from data file? • Issues: How read socket data? Will it affect DAQ? • Process data • Use only released version of reconstruction software • Online Reco in CVS • Offline reconstruction code used to process data but then histograms specific to Online Reco will be produced • Initial version of this code exists thanks to Mark Rayner • Application: OnlineReconstruction • Iterative process – initial version only has TOF information • Create plots • Use Root – Parallel processing on the different online reco computers • PROOF determined to be good way to accomplish this – Paul Kyberd • Save data – Henry Nebrensky • Online Reco root files will be saved along with the run data and the Online Monitoring files • OnRec.####.root Linda R. Coney – 5 Nov 2009
Architecture parallel root (proof) One server distributes events to other cores, and/or other CPUs Automatically returned to server for aggregation. Potential to scale well. Runs C++ code (compiled), so will run the offline code or some (simplified) version. Runs C++ as a scripting language so the possibility of rapid response to new requirements. Integrate the Date reading software during CM25. Get a test process running on old data, during the week 7-14 of November. 23-24 November dedicated runs to test system. Around a week later second run to fix problems. Initial version running from 7 December Solution Rationale Schedule Linda R. Coney – 5 Nov 2009
Online Reconstruction Questions • Do we need absolute real-time information? Is one-file delay a problem? • Which calibration to use? • Whatever is in the release. • How many plots do we want? • Corrolary: What is the processing overload? • Ntuples with the ability to play randomly with the data is not intended. • Do we use already existing beamline design tools (ex. Chris Rogers)? • Will it be done by the end of November? • No. We will have a simplified version to show that the method works and will produce TOF plots. • Iterative process – add more as more reconstruction capability exists. • ad Linda R. Coney – 5 Nov 2009
Conclusions • Now: • Read out and decode DATE DAQ from MICE beam data • Can reconstruct TOF, CKOV, Tracker data • Initial version of Online Reco Application exists • Parallel processing solution found: PROOF • Next: • Implement online reconstruction for Step I • Goal to see first plots in MLCR by end November • Can practice with old data • Start with TOFs and CKOV • Gather more requests for Online Reco plots • Eventually: • Include necessary information for further steps • Routinely have shifters monitoring detectors and MICE physics in MLCR Linda R. Coney – 5 Nov 2009
MICE Online • So far: • DAQ front end • Trigger • Event Building • Controls and Monitoring • Given that we are successfully running the experiment and creating data • How do we know the equipment is working well? • How do we check the data quality? • Two levels of real-time data quality checks • Online Monitoring • Look at raw data for each board in the DAQ • No translation into physical quantities • Online Reconstruction • Initial look at analysis variables and detector information with geometry • Next: see Data Flow MICE notes 252 & 255 Linda R. Coney – 5 Nov 2009