BFC timing [revisited]
BFC timing evaluation for the following options :
Summary : cpu vs. Nglobals

Breakdown by chain options :

Preliminary comments :
####################################################
BFC timing evaluation for the following SSD configurations :

Breakdown by chain options (config 1 and 3 only)

Comments :
config. 1 adds more or less no additionnal time to the event ; it makes sense since config 1 only defines 60 new Sti layers
config. 3 is off-the-charts ... in this case, StiSstDetectorBuilder adds 18140 new Sti layers, then the huge increase in cpu time
conclusion : even config. 3 is out.
- TPC tracking only
- TPC tracking + PXL geometry loaded (StiPxlDetectorBuilder)
- TPC tracking + PXL tracking (StiPxlDetectorBuilder + StiPxlHitLoader)
- TPC tracking + PXL tracking (StiPxlDetectorBuilder + StiPxlHitLoader) + IST geometry loaded (StiIstDetectorBuilder)
- TPC tracking + PXL tracking (StiPxlDetectorBuilder + StiPxlHitLoader) + IST tracking (StiIstDetectorBuilder + StiIstHitLoader)
- 2k hijing events : /star/data99/reco/AuAu200/hijing_382/b0_20/minbias/y2014a/gheisha_on/P14ih/rcf14020_'$3'_50evts.event.root
- run over .event.root files with the chain : "in,y2014a,AgML,usexgeom,FieldOn,MakeEvent,ITTF,Sti,NoSsdIt,NoSvtIt,Idst,BAna,l0,Tree,logger,genvtx,tpcDB,bbcSim,btofsim,-tags,emcY2,EEfs,-evout,-dstout,IdTruth,big,-McEvout,MiniMcMk,-StiPulls,ReadAll,clearmem"
- chain options changed according each configuration to include PXL,IST ,etc ...
- starnew library
Summary : cpu vs. Nglobals

Breakdown by chain options :

Preliminary comments :
- the upper band around cpu ~70[s] is the 1st event processed in the file, as shown here ; it takes more time due to BFC initialization
- primaries are not shown because I forget to select the best (ie highest rank vertex) so the plot cpu vs. primaries is not as clean as globals (but shown here)
- a slight increase (30 --> 40) when adding the PXL geometry
- another slight increase but less pronounced (40 --> 50-60) when PXL hits are used in tracking
- another increase when adding the IST geometry only (50-60 --> 70,80)
- another increase when IST hits are used in tracking (80+)
- StiTpcDetectorBuilder creates 44 layers
- StiPxlDetectorBuilder creates 2 layers (silicon) + 921 layers (supports : PSTM, DTUH)
- StiIstDetectorBuilder creates 4671 layers (including the 1 layer (silicon))
- StiSstDetectorBuilder creates 60 layers (including the 1 silicon layer)
####################################################
BFC timing evaluation for the following SSD configurations :
- config 1 (CVSROOT version) : inactive materials are grouped in 2 mother volumes
- config 2 : mother volumes of inactive materials are breakdown
- config 3 : config 2 but with SFLA, SFLB,SFLC volumes removed (kapton cables, 0.007 cm thickness but accounting for ~87% of the total number of volumes)

Breakdown by chain options (config 1 and 3 only)

Comments :
config. 1 adds more or less no additionnal time to the event ; it makes sense since config 1 only defines 60 new Sti layers
config. 3 is off-the-charts ... in this case, StiSstDetectorBuilder adds 18140 new Sti layers, then the huge increase in cpu time
conclusion : even config. 3 is out.
Groups:
- bouchet's blog
- Login or register to post comments