using sniper in besiii...

Using SNiPER in BESIII analysis

XIA Xin, LI Teng

Shandong University

June 6 2016

XIA Xin, LI Teng June 6 2016 Using SNiPER in BESIII analysis 1 / 21

Outline

Motivation

Data Model

Particle identification and Vertex fit

Performance Test

Summary


Outline

Motivation

Data Model


Performance Test

Summary


Motivation

1 Rapidly increased data samples in BESIIIWith time increases, statistics increases rapidly in BESIIIexperiment. Faster data processing method will help a lot.

2 SNiPER is light-weighted, effective and flexibleFrameworkSNiPER has been used in JUNO experiment in recent years,now also in LHAASO. It is another choice for physics dataanalysis.

3 Analysis the BESIII data under SNiPER frameworkOptimize data structure with Smart Reference.Transplant the BESIII analysis codes to SNiPER.

4 Preparing for the parallel computing in the near future.

BOSS version: 7.0.0


Outline

Motivation

Data Model


Performance Test

Summary


A BESIII analysis workflow

1 Data Service retrieves data from DST via RootEventData2 PID will determine the type of particle for every track3 Vertex fit calculate the vertex position with informations of

the magnetic field and primary vertex from database.


Data model in BOSS

In BOSS, DST files are stored in the form of TTree. All the classesin branches are derived from TObject.

Take data files as example, two TTree are included:1 JobInfoTree: BOSS version, job options etc.2 Event: all the informations of physics events,

part interested.

Data DST

JobInfoTree Event

TEvtRecObject

TDstEvent

TEvtHeaderWhen user read the files, TObject will be converted to theEventObject by Gaudi.


New data model under SNiPER

1. RootEventData is migrated to SNiPER fordirect access to the TObjects in DST.2. Use corresponding TObjects instead ofEventObject.3. Additional branch named DstHeader is added,consists of

1 Parameters for fast pre-selection,2 Smart reference to the other trees.

4. Access to data with DataStoreMgr.

New DSTDstHeader

TEvtRecObject

TDstEventTEvtHeader

Smart Reference provides reference from header to events.Pre-selection can be done with header itself for lazy-loading.


DST conversion

Since no big changes in structure, the conversion will be easy.1 ReBuildDST is developed to convert the BESIII DST to theSNiPER one.

• Read the old DSTs in the normal ROOT way,• Add parameters to DstHeader,• Write to new DSTs with DataStoreMgr, it will do the rest of

jobs automatically.2 Also provide a Service BesInputSvc to access to the BESIII

DST directly.• Parsing the DST file list,• Some useful functions to retrieve informations from files.


Outline

Motivation

Data Model


Performance Test

Summary


Particle identification (ParticleID)

ParticleID is the process of using informations left by oneparticle passing through the detectors to identify the type ofparticle.

• Using dE/dx, TOF, TOFQ, Emc and Muc system

1 Calculate probability for every physical track2 In BOSS, EvtRecTrack is a logical track connecting the

sub-detectors3 EvtRecTrack is the input for PID.


ParticleID in SNiPER

The main problem in new framework is how to get the correctsub-detectors information in one track.

• Need EvtRecTrack to manager the sub-tracks• Initialized from TEvtRecTrack and current scope

• TEvtRecTrack stores the trackIDs of sub-detectors in onereconstructed track

• Scope is used for getting informations like run number

• Connect the sub-tracks with trackID• Use TObjects directly in ParticleID code• Tool AbsCor is added to correct the energy of EMC Shower


Data flow

For analysis, the data flow looks like


Vertex fit

Vertex fit computes the best estimate of the vertex parameters(such as position) with a given set of tracks.Necessary parameters in VertexFit class,

1 Several MDC tracks after PID,2 Primary Vertex information from database,3 Magnetic field information.

To be used in SNiPER,

1 New constructors of WTrackParameter are added2 Run number is passed to VertexFit for database accessing3 MagneticField and DatabaseSvc are detached fromGaudi, rewrote as pure C++ class


Outline

Motivation

Data Model


Performance Test

Summary


Performance Test

• Tested with the analysis e+e− → π+π−J/ψ, (Zc(3900))• Data sample at CM energy of 4.26 GeV, 24 files used in total.

NO. Selection BOSS SNiPER1 charged tracks 4771222 47712222 nGood==4 10028 100283 bnGam>=1 1352 13524 Pass pid 1274 12745 Pass 4C 252 2526 save result 27 277 cms energy 4.258 4.258

Table: The number of selected events under these two frameworks

The selected results are exactly the same.Further test is comparing the physics results.


Mass spectrum comparison

]2) [GeV/cψM(J/

3 3.02 3.04 3.06 3.08 3.1 3.12 3.14

En

trie

s

0

5

10

15

20

25

BOSS

SNiPER

]2) [GeV/c-π+πM(

0.2 0.4 0.6 0.8 1 1.2 1.4

Ent

ries

0

5

10

15

20

25 BOSS

SNiPER

]2) [GeV/cψJ/+πM(

3.2 3.3 3.4 3.5 3.6 3.7 3.8 3.9 4 4.1 4.2

En

trie

s

02468

1012141618202224

BOSS

SNiPER

]2) [GeV/cψJ/-πM(

3.2 3.3 3.4 3.5 3.6 3.7 3.8 3.9 4 4.1 4.2

Ent

ries

0

2

4

6

8

10

12

14

16

18BOSS

SNiPER


Performance comparisonC

PU

[/%

]

20

30

40

50

60

70

80

90

100

Total running time: 1212.5s

Red

iden

t M

emo

ry U

sag

e [M

B]

0

20

40

60

80

100

120

140

160

180

Time usage [s]0 200 400 600 800 1000 1200

Vir

tual

Mem

ory

Usa

ge

[MB

]

650

700

750

800

850

BOSS

CP

U [

/%]

10

20

30

40

50

60

70

Total running time: 758.0s

Red

iden

t M

emo

ry U

sag

e [M

B]

20

40

60

80

100

120

140

160

180

200

Time usage [s]0 100 200 300 400 500 600 700

Vir

tual

Mem

ory

Usa

ge

[MB

]

150

200

250

300

350

400

SNiPER


Time consuming

Tim

e u

sag

e [s

]

600

700

800

900

1000

1100

SNiPER

BOSS

Test NO.0 2 4 6 8 10 12 14 16 18 20

Imp

rove

men

t

0.380.390.4

0.410.42

• Tested 20 times, with 20 different files each time.• SNiPER is about 40% faster than BOSS.

Improvement = (TBOSS − TSNiPER)/(TBOSS)XIA Xin, LI Teng June 6 2016 Using SNiPER in BESIII analysis 19 / 21

Outline

Motivation

Data Model


Performance Test

Summary


Summary

1 Designed new data model for BESIII DST, which can beused under SNiPER.

2 Migrated the ParticleID and VertexFit to SNiPER forphysical analysis.

3 Tested with a real analysis, the SNiPER version workedproperly, and can be 40% faster than BOSS version.

Thanks for your attention!


using sniper in besiii...

Documents