Detector Simulation Meeting
                         Apr 01, 1999 at 4:00pm
                          B0 pump room

                                Agenda

1. news and announcements
2. status of the detector simulation executable (Pasha)
3. TOF simulation - first code in the repository (Tianjie)
4. towards May's release: validation of MC (everybody)
5. other issues

T.V. link is available:

1. news and announcements

• last week Fermilab was hosting Root workshop.  Many experiments are already using ROOT  for different online and offline applications and so far people  who are using  it are pretty happy with the provided tools .
• new  3rd party debugger  - totalview -available on SGI.  Dmitry and Pasha  have been using (and simultaneously evaluating it) during the last weeks.  It is less powerful than  GNU GDB but definitely better than nothing and than KAI-hacked version of GDB for KAI_3_3.  It has light (X-based, no Motif) GUI, no command line interface.  Features which are not supported (or we didn't find how to activate them)

setting up conditional breakpoints

attaching commands to breakpoints

calling object member functions

• invoking totalview:

>setenv TVROOT  ~wike/totalview.3.8.1-1/irix6-mips

>setenv PATH ${PATH}:${TVROOT}/bin

>setenv MANPATH ${MANPATH}:${TVROOT}/man

>totalview    your-program

• with  totalview debugger available  there is no need  to struggle with EGCS any longer, so MC group discontinued support of EGCS
• ROOT-based event display has been moved to development, to build the test executable

>addpkg -h RootMods

>make RootMods.evd

• test example will be provided soon

    several technical  aspects have been discussed:

performance: the last number is 338 sec/100 minimum bias events (MBR) , this is consistent with all the last tiiming measurements (about 12-13 seconds per ttbar event) and means that somewhere there is a slowdown factor of about 1.5-2 we lgot since last spring

profiling:  e5 has been run with the following  TCL file  How to run profiller:

>ssrun -v -usertime bin/IRIX6-KCC_3_3/e5.exe gnt_e5/test1.tcl

>prof -gprof -h -l -quit 1% e5.exe.usertime.18490 > e5.time

looking at the timing block diagram

 

 
 
 

one can make several observations

simulation code is spending more than 80% of all the time in GEANT transport routines  (including GUSTEP)

about 30% of overall time goees to tracing the electrons - where these are coming from?

CDF digitization takes a little bit more than 10% of overall time, so digitization routines do not affect the performance

who calls GTNINO?

Items to be put on the job list:

how different subdetectors contribute to overall timing ?

how the overall time breaks down for different input data, say ttbar events?

---

 

2. TOF simulation (Tianjie)

• the postscript version of the talk can be found here
• proposed format of TOF data bank

physics validation of the MC is practically missing. The proposed solution is to create a specialised histogramming module (prototype - RootMods/TGeant3Module.cc) which would do histogramming for the MC. No histogramming should be done inside the simulation code unless  it is really necessary .

we have working reconstruction code for tracking, calorimetry and IMU. Need to explore it for validation purposes

Pasha volunteered to write a base class module which would be oriented on using ROOT histogramming tools and which  functionality should include

• opening/closing ROOT files
• creating a directory per subdetector to store histograms
• creating a higher-level subdirectory (per module)
• proper changing the directory  to avoid clashes between the modules and subdetectors
• interactive plotting of the filled histograms

a name for such a class has been discussed earlier with Marjorie and Liz Sexton, RootHistModule seems to be an appropriate name for such a class

George raized an issue of using HBOOK for histogramming. Technically this is also possible, one can either use HBOOK directly from C++ ( using cern_i/hbook.h interface) or to use HepTuple. However if people are using both ROOT and HBOOK within the same job,  at least 2 different histogram files will be produced in the end. HBOOK file can be converted into ROOT format using h2root utility

George also asked if it is possible to convert PAW "almost-FORTRAN" COMIS files into ROOT "almost-C++ " scripts. Pasha suggested using f2c , but  the experiment performed right after the meeting showed that it is much easier to do it manually . The conversion requires certain work but technically is pretty much straitforward. There is no easy way known to convert KUIP macros into C++.