150 likes | 245 Views
GTT results from recent ep runs. Credits Run configurations Rates and latencies Data sizes CTD+MVD versus CTD only operation Algorithm performance Typical events Conclusions and next steps Important statements are rendered red and points of interest in green. Credits.
E N D
GTT results from recent ep runs • Credits • Run configurations • Rates and latencies • Data sizes • CTD+MVD versus CTD only operation • Algorithm performance • Typical events • Conclusions and next steps Important statements are rendered red and points of interest in green. GTT results from ep runs 42311-42318 (CY)
Credits • DAQ infrastructure. • A.Polini a, M.Hayes b* and C.Youngman a. • CTD+MVD tracking algorithm. • M.Sutton c, B.West b*, R.Hall-Wilton b, J.Loizides b,e and B.Straub f. • CTD interface. • S.Topp-Jørgenssena. • STT tracking algorithm. • M.Soaresd* and D.Gladkovg. • STT interface. • H.-P.Jakobg and A.Stifutking. • GTT farm and network hardware. • S.Dhawanf. • * now working on other things …. • a) DESY, b) UC London, c) Oxford, d) York Toronto, e) Argonne, f) Yale and g) Bonn. GTT results from ep runs 42311-42318 (CY)
Run configurations • Runs taken 42311 - 42318, from 13/9/02 with • Ie 9.8 - 8.3 mA • Ip 9.5 - 9.2 mA • CTD at 100% voltage (42311 at 95%) and MVD on (42311+42312 off) • PHYSICS, STD_020814_HIGH and ALL-BAC • GTT algorithm analysing CTD+MVD data • STT data available but not sent to GTT • Comparison run 41188 from 21/6/02 (IC collaboration meeting) with • Ie 4 mA • Ip 20 mA • CTD on 100% voltage and MVD on • PHYSICS, STD_020621_2_HIGH and ALL-BAC-LUMI • GTT algorithm analysing CTD+MVD data • STT data not available GTT results from ep runs 42311-42318 (CY)
Rates and latencies • Performance of GTT has significantly improved w.r.t. 41188 • Mean GSLT latency is approaching acceptability. • GTT CTD only mode excellent. 2ms slower w.r.t. CTD-GSLT, but using stereo not z-by-time !! • GTT CTD+MVD mode slower by 6ms (~2 decoding and ~4 cluster transfer) • Deadtime reduced. • CTD only no additional deadtime • CTD+MVD increases to ~9% - transfering to much cluster data, see later. • Transfer of Strip data has no effect on rates and latency • Reasons: harder GFLT trigger and smaller CTD+MVD data sizes at GTT. • Next 2 slides take quick look at the data sizes – they´re quite different. GTT results from ep runs 42311-42318 (CY)
Data sizes run 42314 A B • Data size/event log10(bytes) • A – MVD strip • B – CTD • C – CTD+MVD cluster • D – MVD strip+cluster • Mean size • A – 17.4 KB • B – 0.8 KB • C – 10.3 KB • D – 28.6 KB C D GTT results from ep runs 42311-42318 (CY)
Data sizes run 41188 A B • Data size/event log10(bytes) • A – MVD strip • B – CTD • C – CTD+MVD cluster • D – MVD strip+cluster • Mean size* • A – 31.0 KB • B – 7.2 KB • C – 45.4 KB • D – 75.3 KB • * At IC most likely size was used. C D GTT results from ep runs 42311-42318 (CY)
CTD+MVD versus CTD only operation • Left GTT CTD+MVD mode (42314) and right GTT CTD only mode (42315) • Top MVD0 crate (upper barrel) cluster data size/event 1 to 10KB range. • MVD0 4KB offset, MVD1 (lower barrel) 1.5KB and MVD2 (wheels) 0.0KB. • Offset results from noisy modules,Removing their ~4-5KB will improve deadtime and latency. • Middle MVD decoding time. • ~2ms mean but long tail seen only in log plot • Bottom Z vertex • Addition of MVD data into track fit pulls final vertex strongly as MVD often busy excess at -80cm washed out. • What is the bump at +150cm ? • Should use MVD data only when sensible. Cutting on data size at front-end ADC and not sending to GTT means, that the cluster data will not be complete offline, but we´re sending strip data anyway. Acceptable ! GTT results from ep runs 42311-42318 (CY)
Algorithm performance • DIS data sample event requirements • From runs 42311 – 42318 where MVD readout is OFF • > 4 tracks reconstructed • 40 < E – Pz < 65 • 1 e prob. > 0.9 • All data sample event requirements • Same runs as DIS sample • > 4 tracks • Remember the algorithm results come from runs with no MVD input. • Look at difference in ONLINE GTT and OFFLINE simulation GTT results from ep runs 42311-42318 (CY)
DIS selected events A B • Tails in A possible as the simulation data cutoff size is not yet correct. • Corrolation in B shows good agreement of simulation with online algorithm. • C shows that cutting on the CTD data size would not loose many events. • D tail related to A‘s or different ? C D GTT results from ep runs 42311-42318 (CY)
DIS selected events A B • A+B show z vertex distributions of online and offline simulation. • Events with no TLT vertex (plotted as 0) can be seen as the horizontal band in C. • Difference in entries A-D is the number of events with no TLT vertex – why 22/250 ~ 10% needs confirming. C D GTT results from ep runs 42311-42318 (CY)
All events A B • C – many badly reconstructed events associated with large CTD occupancy C D GTT results from ep runs 42311-42318 (CY)
All events A B • A+B+D show the known z vertex features. • Why the band at -80 cm exists, in D, is not understood. • Difference in entries D-A is the number of events with no GTT vertex 3343/16903 ~ 20%. Why – are these events with secondary vertex ? C D GTT results from ep runs 42311-42318 (CY)
Typical events viewed with GTT display Run 42314 Event 938 Run - Event - GTT results from ep runs 42311-42318 (CY)
Typical event viewed offline GTT results from ep runs 42311-42318 (CY)
Conclusions and next steps • Improvements in GTT latency at GSLT and deadtime at GFLT, driven by harder GFLT trigger and better beam/vacuum reducing the data sizes, are encouraging. Have reached latency and deadtime design expectations in CTD only mode ! Well for this GFLT accept rate. • Understanding the algorithm performance (comparing online and offline simulation, etc.) is producing interesting results associated with debugging and w.r.t. other tracking components. • Next steps: • Two GTT trigger requirements now included into GSLT • -20 < Zvertex < +20cm • CTD data bytes < 2000 Bytes • Want to monitor GTT decision at GSLT and veto some ep runs. • Continue understanding the algorithm. • Improve data size (remove noisy MVD modules) and possible front-end cuts. • Use playback system to study latencies. • Put STT back in. • Etc. • GTT status is not hopeless but depends very much on data sizes. GTT results from ep runs 42311-42318 (CY)