150 likes | 257 Views
Online Data Challenges. David Lawrence, JLab Feb. 20, 2014. Online Data Challenge 2013. Primary participants: Elliott Wolin , Sean Dobbs, David Lawrence. When: August 26 – 29, 2013 Where: Hall-D Counting House
E N D
Online Data Challenges David Lawrence, JLab Feb. 20, 2014 Online Data Challenges
Online Data Challenge 2013 Primary participants: Elliott Wolin, Sean Dobbs, David Lawrence When: August 26 – 29, 2013 Where: Hall-D Counting House Objective: Test data flow and monitoring between final stage Event Builder (EB) and tape silo (i.e. neither the DAQ system nor the offline were included) Online Data Challenges
Input Data • Pythia-generated events simulated, smeared, and passed through L1 event filter* • Events digitized and written in EVIO format • mc2coda library used to write in the new event building scheme specification provided by DAQ group • Translation table derived from Fernando’s spreadsheet detailing the wiring scheme that will be used *event filter may have used uncalibrated BCAL energy units, but resulted in roughly 36% of events being kept. Online Data Challenges
Computers systems (many of these on loan) *n.b. all L3 machines connected via InfiniBand Online Data Challenges
L3 Infrastructure Test • 10 nodes used to pass events from Event Builder (EB) to Event Recorder (ER) • EB on gluon44, ER on halldraid1 • Two “pass-through” modes used: • Simple buffer copy without parsing (40kHz) • Buffer copy with parsing and application of translation table (~13kHz) • DL3TriggerBDT algorithm from MIT • Unable to run for extended time without crashing • Cause of crashes unknown and under investigation • MIT and TMVA code itself have been eliminated as causes • Total rate of ~7.2KHz Online Data Challenges
L3 Prototype Farm 10 Computers ordered and will be shipped next week for L3 prototype farm Fast Computers used for testing (gluon44, gluon45) Computers ordered for L3 infrastructure Max rate estimate for L3-BDT prototype farm: (1.6kHz)(12366/5104)(10 nodes) = ~39kHz Online Data Challenges
Monitoring System (RootSpy) test • Histograms produced by several plugins were displayed via the RootSpy GUI • Overlay with archive histograms • RootSpyarchiver (writing summed histograms to file) • Integration with CODAObjects • Still need to fully implement final histograms mechanism Post-L3 Monitoring Pre-L3 Monitoring Online Data Challenges
Other monitoring • Ganglia installed and working for monitoring general health of all computer nodes • JANA built-in monitoring available via cMsg allowing remote: • Probing for rates • Changing number of threads • Pause, resume, quit (either individually or as group) Online Data Challenges
RAID to Silo test • Transfer from RAID disk to silo tested • At least 50MB/s achieved, but possibly higher • Certificate and jput set up, but we were informed later that a different mechanism should be used for experimental data from the halls • Will arrange for IT division experts to come run tests and educate us on the proper way we should be transferring to the silo Online Data Challenges
Primary Goals Test system integration from L1 trigger to tape using low rate, cosmic trigger Test system integration from ROC to tape using M.C. data at high rate Test calibration event tagging3 secondary goal: Test multiple output streams *Fully installed and calibrated trigger system is not required (only 1 crate needed). *Fully installed controls system is not required (only on/off of some channels needed). Online Data Challenges
Differences from ODC2013 • Data will come from crates in hall • CODA-component Farm Manager • New ROOTSpy features • advanced archiver functions • reference histograms • High speed copy from RAID to Tape Library • Faster farm CPUs (16 core, 2.6GHz Xeon E5-2650) Online Data Challenges
Schedule ODC2014: May 12-16, 2014 Online Data Challenges
Summary • EB to ER Data flow piece tested • L3 infrastructure tested and works in pass-through mode at 40kHz (mysterious issues with L3 plugin still being tracked down) • Monitoring system tested • Identical pre-L3 and post-L3 monitoring systems • RootSpy GUI used with multiple producers • RootSpyarchiver • RAID to Tape silo tested • Successfully transferred > 1TB from counting house to silo at >= 50MB/s • Rate seemed slower than anticipated by factor of 2, but measurement mechanism not accurate due to staging • Alternate transfer method has been advised and will be pursued Online Data Challenges