processing of 1999 data
DESCRIPTION
Processing of 1999 data. 1999 L dt = 2.42 pb -1 7.7 × 10 6 f ’s collected 1.1 × 10 6 K S K L tagged by K S p + p - 6.0 × 10 5 K + K - tagged by vertex All data reconstructed at acquisition. Analysis executable: CVS source control Development history Version-tagged output. - PowerPoint PPT PresentationTRANSCRIPT
Processing of 1999 data
1999 L dt = 2.42 pb-1
7.7 × 106 ’s collected
1.1 × 106 KSKL tagged by KS
6.0 × 105 KKtagged by vertex
All data reconstructed at acquisition
Analysis executable:
• CVS source control• Development history• Version-tagged output
0 Short data-taking periods in Aug, Oct, Nov ’99 at acquistion (total ~1 pb-1)
1Main data-taking Dec’99 at acquisition (1.4 pb-1)
Aug, Oct, Nov ’99 data off-line (Feb ’00)
2Changes to algorithms in this report (DC reconst.)
Currently underway for all data
Computing resources
Tape library6 Magstar drives15 MB/sec each40 GB/tape(uncompressed)5500 slots220 TB
GIGASWITCH
Fast Eth,GbitSwitch
OFFLINE FARM 10 Sun Enterprise 450 (4 UltraSPARCII, 400 MHz)700 SpecInt95
Offline farm disk server2 Sun Enterprise 3500
Tapeserver
Tape server
SCSI
SCSI
FDDI
FDDI
Fast Eth
SCSIGbit Eth
ONLINE FARM7 IBM H50 (4 PPC 604e, 330 MHz) 420 SpecInt950.5 TB local disk space (SSA)
0.5 TB RAID
Fast Eth
•All data written to disk in 1 GB files (640 K 1.63 KB evts)
•Reconstruction/streaming performed on dedicated farm
•Production starts on-line and follows acquisition
•Single-CPU job turn-around in 4 hrs
Overview of offline reconstruction
Translation
Cluster reconstruction
Absolute event t0
Background filter
DC hit reconstruction
DC track/vertex recon.
Track-to-cluster assoc.
Event classification
KLKS
Cosmic filter
Rad
K+K Bha
Dedicated
DC hit reconstruction
DC track/vertex recon.
RAW
Track-to-cluster assoc.
Calibration Bhabhas
UFO
datarec “simplified” flow diagram
5 ms/evt
100 ms/evt
Typical data stream composition, Dec ’99
L = ~ 1030 cm-2 s-1 Tot. rate ~1.3 KHz
Unvetoed cosmics ~700 Hz
Prescaled cosmics ~400 Hz
Mach. Bkg + Bhabha < 20° ~200 Hz
Physics ( + Bhabha > 20°) ~15 Hz
Background filters
Cosmic ray and machine background filters use complete EmC reconstruction + number of DC hits
Recent changes to filtering algorithms• increase cosmic ray and MB suppression• decrease inefficiency incurred for physics
channels
Cosmic filter
suppression raised from 84% to 97% with decreasing physics losses
Machine background filter
suppression highly variable depending on run conditions• 40-90% over all KLOE runs• 50-60% for Dec ’99 data
Final stateCosmic filter
MB filter
KSKL 0.02% 0.2%
K+K 2.4% 1.1%
0.07% 0.07%
Neutral
Radiative0.3% <0.01%
0.2% <0.04%
Inefficiency incurred for physics channels
Calorimeter reconstruction
?
Basic clustering algorithm:
• cell readout: {EA, EB, TA, TB}i {E, x, y, z, t}i
• {x, y, z, t}clust from energy weighted avg. over cells
• Missing information systematically underestimate Eclust
New analysis module:
• Uses zclust to get attenuation length correction
• Allows EA and/or EB to be summed into Eclust
Improvements to clustering algorithm Improves measurement of , , masses
Correction of TDC calibration constants
1% error on abs. scale for conversion constants (ps/count)
• ~60 ps error on prompt TOF• Should improve accuracy of neutral vertex reconstruction
ee
Drift chamber reconstruction
•Major effort to understand systematics for momentum reconstruction in DC a priori
•Ad hoc prescription available for some time
Drift chamber geometry
Magnetic field map
Energy loss corrections
p vs , Bhabha events
p(MeV/c)
p (
MeV
/c)
ee
KL
KS
KL lKL l
KL
KL
decay in DCdecay at IP
Many event samples studied
Drift chamber geometry
Stereo angles in reconstruction geometry decreased by ~0.5%
Before correction
After correction
Effect on p vs. , Bhabha events:
p (M
eV/c
)
(deg)
Vertex fit includes new accounting of materials around interaction point
DC wall:650 m CF +50 m Al
Beam pipe at IP:Cylindrical
DC wall:700 m CF +200 m Al
Beam pipe at IP: Spherical
OLD:
NEW:
Effect observed:~ 1.5 MeV step in Mmiss(0) from KL decaying
inside DC (p 400 KeV)
Geometrical adjustments to field mapVarious probe alignment errors detected by detailed analysis of trends in field components in raw map:
• 10-50 Gauss in Br, B
• p(= 20) increased by ~0.7 MeV/c (Bhabha events)
r
r
z
z
r
r
r
z
z
Measurement device:
Cross mounting:
6 Hall probes
Rotating arm with
28 crosses
Alignment errors revealed by study of raw field map:
Global rotation of arm
Misalignment of probes on cross
Gravity-induced torsion on arm
Abs. calib. from NMR probe at
r = 0, z = 0
g
Saturation of field map
Previous reconstruction version: Bz(I = 2660 A) × 2500/2660 Bz(I = 2500 A)
Comparison of maps at 4.5 and 6 KG shows saturation effects depend on (r,) and especially z
Bz(I) from NMR probe (r0, z=0) shows non-linearity:~30 Gauss error in abs. scale of Bz
from extrapolation
Corrections to Bz using NMR data and maps at 4.5 and 6.0 KG reduce p() effect to 1 ppt
Original plan was to run with I = 2660 A Bz = 6 KG
Bz(4
.5) –
0.7
5Bz(6
.0) G
auss
z mm
4.56.0~5.6
Effect of corrections
Bhabha events KS
p (M
eV/c
)
(deg)
±0.5 MeV/c = p/p ~ 0.001
Energy loss corrections
Track/vertex fit includes energy-loss corrections in gas/wall using m = m
2nd pass to re-track identified K+K with m = mK in K+K stream
New materials for dE/dx calculation eliminate step in Mmiss(0) vs. rxy for KL
Event classification
Stream Tagging algorithms/channels Retained Events/nb-1 MB/nb-1
raw raw 762000 1260
kpm KK w/w/o vtx, single K± recon 1000 33
ksl
KS: chg vtx in DC, KL EmC, KL , KL 3
KL: KS , KS
KS lrecon 6700 190
rpi , (prompt chg vtx w/ KS veto) recon 270 6
radneutral radiative decays (prompt neutral energy in EmC)
charged/neutral radiative decaysrecon 720 13
clb , recon 60 1
bhaGolden Bhabhas (5.7 b)
Large-angle Bhabhas (0.75 b)recon 5700 91
flt All events passing FILFO raw 112000 180
afl 1/100 of events rejected by FILFO raw 4200 6
cos 1/10 of prescaled trigger cosmics raw 23000 33
Event streaming1.3 KHz
900 Hz
raw
EmC recon.
DC recon.
DC recon. Evt. Class
bha
kpm
ksl
rpi
rad
clb
flt
afl
190 Hz
7 Hz
Evt. Class
cos
prescaled cosmic
40 Hz
MBcosmic
÷10
÷100
Bhahba 15 Hz
10 Hz
200 Hz
Rates assume typical Dec’99
running conditions
Rates and code optimization
Algorithm
DBV-1 DBV-2
CPU/call (ms)
CPU % of job
CPU/call (ms)
CPU % of job
Overhead
Translation3.1 12.4 2.9 14.2
EmC recon.
Global t01.3 4.9 2.2 10.3
FILFO
Prescale filter0.4 3.3 0.3 3.8
DC track fit 110 67 75 55
Other DC recon. 8.0 4.8 10.4 7.4
Track-to-cluster 8.4 4.9 9.9 7.0
Event classific. 1.5 0.9 2.1 1.6
Diagnostics 3.9 1.6 1.0 0.5
Total Rate 40 Hz 48 Hz
Throughput on 40 CPU offline farm:
Dec ’99 data: 1900 Hz (DBV-2)
L = ~ cm-2 s-1
DC trigger, prescaled cosmics
Aug ’99 data: 2400 Hz (DBV-2)
L = cm-2 s-1
no DC trigger, no prescaled cosmics
Work started on CPU optimization
Changes extensive in online reconstruction (monitoring)
• Throughput increased by factor of 3!
Some optimizations propagated back to offline reconstruction (work in progress)
• DC track fit 46% faster Reconstruction chain 20% faster
OFFLINE
root hist. server
Trigger monitortrigger performancebackground rateluminosity estimate
Offline monitoring: W, , p
L3 spies
Bhabha, CosmicMIP
EMC monitort()E(Bhabha)MIP
DC monitorcell effic.residualsIP, p monitor
root browserillumination
Event display
raw
L3
Calibration
DANE
DANE
Online calibration and monitoring
SWITCH
BUILDER
KID
Drift chamber online calibration
EmC recon
selective filterraw
DC tracking
residuals
DC CALIBGO
STOP
DC CHECK • starts automatically every run • integrates 300K cosmics (3 hr)• histograms track-hit residuals• 50 m residual tolerance 8%
100% = 400 Hz
selcos raw32 Hz
HepDB
DC CALIB• reconstructs selected evts using
residuals (45 evt/sec, ~2hr)• fits s-t relations • stores new calibrations in DB
along with DC conditions
Implemented at script level
All reconstruction proceeds with residuals < 50 m for upcoming data taking
DC CHECKOK
Calorimeter online calibrationMIP-cosmic run
vfib, t0, t0, MIP response
24 hrs, every 30-60 daysPrescaled cosmics
monitor t0
online, every run
eventsmonitor gbl t0, update DB
online, every run
STOP
eventsfine t0 adj by column0.5 hr, every 100 nb-1
Bhabha eventsfine equalization by col.update HepDB, online
eventsAbs energy scale
update HepDB, online
GO
100 nb-1100 nb-1
t0 shift
Timing Energy
no
yes
Online reconstruction monitor
Fast versions of reconstruction algorithms run on-line for monitoring
Bhabha tracks extrapolated to z-axis measure:• position () and size (L) of luminous region• machine boost (p)
Values written to DB, available for analysis
L3 Bhabha(+)
L3 cosmic
EMC monitorEtot, Ecl
Tcl, TclR/c, TclL/vE
EmC vs. EDC
for ee
DC monitorcell efficienciestrack-hit residualsIP and boost: , L, p
EmC + DC reconstruction
EmC + DC reconstruction
65 Hz
Reconstruction and quality controlRejected rates:
MB, cosmics, Bhabhas
Bhabha rate:
L dt, L
Event rates, :
KS , , KL EmC, , KK
Bhabha:
W, p, Etot, E/p, TL/c
KS:
Minv, pK, 2(dN/dcos)
KK:
Minv, p(K), p(K)
Many variables continuously monitored during data processing
Web interface
Graphical history interface
Monte Carlo productionPlan to generate and reconstruct ~11M events
Production environment similar to that used for reconstruction:
• Same executable used for official reconstruction • Output files are version-tagged, have DB entries
MC production runs on offline farm, or on new Linux farm (to be acquired soon)
Work to be completed:Not conditioned on data:
• precise reconciliation of EmC and DC geometry• introduction of new generators
• BABAYAGA (Pavia): Bhabha generator with radiative corrections
• EVA (Karlsruhe): ee generator with ISR+FSR
Conditioned on data:• new field map if next run at IB = 2300 A• finalization of physics program
Event type Request
all 5.0M
KSKL
KS ; KL
KS ; KL charged
KS ,e
1.9M
KK
K ,e0.5M
radiative channels,,fa
ee
3.1M
calibration channelseeee
ee
cosmic rays
0.3M
Total ~11 M
Conclusions
• KLOE reconstruction has been thoroughly proven on all fronts:
algorithms, procedures, environment, and monitoring.
• Emphasis while waiting for luminosity is on refinements.
• New online calibration procedures for upcoming data-taking.
• Next step:
Monte Carlo production for studies of efficiencies and systematicsdriven by requests from analysis groups.