Offline Status ReviewM. Moulson, P. Valente, for the Offline Group16 March 2001

Outline:• Status update• FILFO: new developments (G. Finocchiaro)• Questions to discuss• MC production on Linux (P. Valente)

Processing of fall 2000 dataReference data set:

Runs 15197—17330 (23 Sep—11 Dec)5.61G triggers, 15543 files, 13.6 TB raw data

Offline L dt = 23.1 pb-1

• DBV-6: 5.5 pb-1

• DBV-7: 18.0 pb-1

Total events collected:• 130M Bhabhas• 67M decays

est. = 2.9 b

• 10.9M KL tags

• 7.2M KL crash

• 19.5M KK w/ vertex• 6.2M tags

Event streaming2050 Hz

1650 Hz

raw

EmC recon.

DC recon.

DC recon. Evt. Class

bha

kpm

ksl

rpi

rad

clb

flt

afl

400 Hz

12 Hz

Evt. Class

cos

prescaled cosmic

40 Hz

MBcosmic

÷10

÷100

Bhahba 40 Hz

50 Hz

450 Hz

Typical raw data composition, Nov-Dec ’00L = ~ 1030 cm-2 s-1 Tot. rate ~2050 Hz

Unvetoed cosmics ~ 700 Hz

Prescaled cosmics ~ 400 Hz

Mach. Bkg + Bhabha < 20° ~ 900 Hz

Physics ( + Bhabha > 20°) < 100 Hz

Processing statistics for fall 2000 dataRaw data: 5605M Events, 13.6 TB

Physics: 2.0% of total events25% of raw volume160 GB/pb-1

Bhabhas: 2.3% of total events20% of raw volume114 GB/pb-1

Stream Events (M) Volume (GB) Summarykpmkslrpiradclb

19646

231

687227016652019

112M Events3.6 TB

bha 127 2623fltaflcos

12483196

379192273

1379M Events4.2 TB

Avg. fraction ofevents tracked:

26%

Avg. bandwidth on 40 CPU’s: 1060 Hz

Avg. time/trigger

38 ms

Avg. time/tracked evt.

151 ms

datarec performance in fall 2000Daily averages

Average uptime efficiency: 63%Staging and procedural losses count

Evts. acquired Evts. processed

Once in stable situation, able to follow with ~75% of data processed

No

DC

trig

End

of d

ata

Follo

w-u

p

Afte

r Xm

as

Mis

sed

runs

Rep

roce

ssin

g

Acquistion rateReconstruction rate

Farm bandwidth

23 Sep 11 Dec

Brief history of datarec executableDBV-4 First version of datarec with all tracking upgrades complete

Used for 1999 MC production

DBV-5 Event classification sub-flags modifiedProblem discovered:

All events with 2 and >1 thrown away

DBV-6 DBV-5 bug fixedSub-flags for rpi, clb streams modifiedVery large angle Bhabha counter addedProblem discovered:

FILFO Bhabha filter causing losses in neutral radiative channels

DBV-7 DBV-6 condition for running FILFO Bhabha filter fixedLAB selection cuts modified (Emax)

New version of datarec executable

Already in DBV-8:

CLUFIXENE Now runs on MC events Calibration factor adjusted

EMCALIB Slewing corrections added

VTXFIN Path length correctedOther minor bugs fixed

(Almost) ready for inclusion in datarec:

FILFO Fast tracking cutsNhits definition excludes pileup

A_C zlib event compression

FILFO counter addedFills ECLS for rejected eventsCan write ECLS w/o ECLMOD

TRCLMOD Track-cluster projection fixed

ECLMOD KS semileptonic tag added tag modifiedNew ECLS bank version

VLAB New luminosity module

New version of GEANFIGEANFI version 1.06/07 released on 6 Feb:• EVA generator inserted

generator with different models for ISR, FSR, interference with f0

• decay inserted• decay (C-violating) inserted• Improved simulation of K 3 decays generator

added parameter to specify fraction of decays that don’t proceed via • Modifications to GEANT tracking to allow generation of specific KL topologies

KL decaysin beam pipebefore entering DCbefore entering calorimeterin calorimeter

• Change name of output banks: DTCE MDTC, CELE MCELMakes it easier to re-reconstruct MC files that have already been reconstructed

• New default value for magnetic field: I = 2300 A

Status of MC production

Event type Request Today Added

all 2.4M 2.4M

KSKL

KS ; KL various

KS ; KL all (1.1M)2.9M 4.8M

Mostly KS , KL

Some KS KL anywhere

KS semileptonic, KL all

KK

K ,e0.24M 0.24M

radiative channels,,fa

ee

2.0M 2.6M Mostly events

calibration channelseeee

ee

cosmic rays

0.8M 0

Total ~8.3M 10.0M

MC production for 2000 data started 9 Feb

What needs reprocessing?Reprocess DBV-6 (5.5 pb-1) for neutral radiatives?

FILFO Bhabha filter deleterious to neutral radiative eventsRejects 40% of events

Pre DBV-7: Only runs on “charged” eventsCharged event: Nhits > 30Pileup contribution to Nhits in 2000 data

DBV-7: Bhabha filter vetoed by neutral radiative tag instead of Nhits cut

Luminosity to 1% only possible with > DBV-7 reconstruction from raw

Reprocess from raw w/o trackingRetain only rad, bha streams

Reprocess DBV-6, 7 (23.1 pb-1) for tracking updates?

Track length, TCA projection corrections can be handled at analysis stage

Wait on DC calibrations for datarec? s-t constants sensitive to pressure

Calibration in 2000 when residuals > 20 m

Mean frequency of calibration: 1/week

Observation:

Large changes (~100 m) in residual distributions observed only after hardware interventions

Proposal:

50 m < residuals < 100 mEstimated loss in VTX = 1%Warning to calibration operatorsReconstruction not stopped

Residuals > 100 mStop reconstructionWait for calibrationEssentially protects against human error

Atm

osph

eric

Pre

ssur

e

Wait on EmC calibrations for datarec? Two regimes:

100 nb-1 in ~hours:Easy to wait, don’t have to wait long!When calibrations not available, plenty of data on tape to reconstructRequires stable & reliable calibration processesfluctuations > calibration: does it matter?

100 nb-1 in ~days:May have to wait a long time!When calibrations arrive, they have limited validity: fluctuations < calibration

These runs typically not used for analysis

Intermediate regime:Could be where waiting matters most

Available flags:

CT0G TRF, cable

events, every run

CABT fine t0 adjust by column events, every 100 nb-1

CABE Abs. E scale/column equalizationBhabha + events, every 100 nb-1

Other considerations:

Observation from last 4 weeks of 2000 data:Use of calibrations from previous 100 nb-1 usually resulted in t0 within tolerance

EmC reconstruction gets redone for Ntuples and DST’s in any case

Currently evaluating impact on streaming

Retain flt and afl streams?

flt stream (FILFO-selected events) • Adds ~25% to raw data volume (3.8 TB on top of 13.6 TB)• Has not been seriously used to date• Potentially useful for recovery from tracking problems• Possible solution: delete flt files after the fact

afl stream (1% of FILFO-rejected events)• Has not been seriously used to date• 75% of raw events are FILFO-rejected, FILFO is fast• Negligible benefit to keeping afl stream in datarec chain

How to handle prescaled cosmics?Trigger cosmic veto known to cause ~50% inefficiency for events with high M2()1/5 of vetoed cosmics retained by trigger to study bias (“prescaled cosmics”)

Current situation10% of prescaled cosmics retained as raw events in cos streamSame 10% passed through reconstruction chain, events end up in rad stream

New proposalNew code (P. Branchini) selects cosmic vetoes that could be events

Efficiency: No losses on ~1000 cos events that pass filterRejection: 93% rejection of uncut cos event sample

Incorporate in selcos module for an intelligent software downscaleEntire sample of cos events ends up in rad stream

Useful to retain unbiased cos stream for other analyses?Effect of cosmic veto on KK < 1% (M. Martemianov, Kaon WG)Expected effect for KSKL events minimal

p spectrum of ’s that trigger comsic veto dies out below 300 MeV

What information do we keep in DB2?Counters: VLAB, Gold Bhabha, , KS, KS, KL crash20 fields KK, , Neutral Radiative, Rejected Bhabha, cosmic, MBJob table +8 unused fields

Fitted quantities: s from Bhabhas, Etot from EmC, TL/c from Bhabhas20 fields M(KS), M(KS)Job table +15 unused fields

Status flags: Biases: Indicates any reconstruction errors that cause biases in stream 2 flags per stream Obsolete: Indicates if file has been superseded by a new reconstruction datarec file table

Encoding of status flags:Word of bits independently set to indicate reasons for bias/motive for reprocessing

Calibrations (DC, EmC t, EmC E)Bugs (FILFO, clustering, tracking, each ECL filter, etc.)

No bits set means stream good to use for analysis

Example KID query:input url "dbdatarec:stream_code=’rad’ and run_nr between 15197 and 17330and Biases=0 and Obsolete=0"

What is needed for MC production off-site?

Obstacles to off-site production:

Inability to read-write connect to DB from outside LNFRead GEANFI cards from DB2Update processing/processed flags, random seedsKeep track of files produced

Relatively easy to enable outside access for individual nodes on request

Scripts that coordinate generation/reconstruction designed for LNF setupNeed modifications and testing to run at remote sites

Need mechanism for archival output files

Proposal for mini-DST production

Mini-DST’s to be produced:• in dedicated production pass• in zlib-compressed YBOS format•with a bank subset essential for physics analysis, including summary banks•for different event classification substreams

Status:zlib compression DoneScripts Some work neededRetained banks Needs definitionSummary banks Needs definition, codingSubstreaming Needs definition

Waiting on physics input—Technical aspects relatively simple

DST substreams: initial proposalInitial proposal based on ksl Ntuple, radiant substreams, datarec streams, and EVCL tags

ks2pi KS , KS

ksemil KS semileptonic decaysklcr KL crashkl2p0 KL

kl3p0 KL

klppp KL

intertag interferometry tagskpm kpm (KK) streamrpi rpi () streamneu_max4g neutral events, at most 4 prompt ’sneu_min5g neutral events, at least 5 prompt ’scha_min3g 2 tracks, 3 or more prompt ’scha_ppg tagged eventsclb clb (, ) stream

DST banklist: initial proposal

Raw Banks (?) 2 KB/evt w/w/o compressionAllow recovery of information not in DST

LRID, EVCL Event headers

T0GL Global t0

TRIG (TPIE, TMON, TCAF, MOKA) Trigger summary Based on Ntuple block, loosely discussed

CTRG Sector-to-cluster

CSPS (CELE, CWRK) Calorimeter hits Almost ready

CLUS (CLPS, CLLS) Calorimeter clusters Planned

QCAL (QCAE, QWRK) QCAL hits Based on Ntuple, CSPS; new to this proposal

DC hit counts Based on Ntuple, new to this proposal

DTFS w/o hit lists Fitted tracks New to this proposal

DVFS Fitted vertices

TCLO Track-to-cluster

ECLS, ECLO Event classification

KNVO, VNVO, INVO Neutral vertices

Mini-DST size estimateExample mini-DST size estimates prepared with help of S. Miscetti

ksl stream: KS , KL events selected by S. DiFalcoOriginal event size: 34.4 KBCPTEST3: 11.4 KB+gzip 7.3 KB+dropping raw ~5.5 KB

Typical DST event size for kpm, ksl substreams: ~7 KBTotal reduction factor not accounting for event selection: ~5

rad stream: substreams from radiantOriginal event size ~23 KBcha_min3g 7.6 KB (+gzip 4.9 KB)neu_max4g 3.6 KBneu_min5g 4.0 KB

Typical DST event size for rad subtreams: 3-5 KBTotal reduction factor not accounting for event selection: ~5

Estimated zlib compression factor: ~1.6