GENIE-MC/Generator

Possible memory leak?

Opened this issue · 2 comments

pedrosimpas commented

Dear GENIE developers,

I tried to run GENIE version 3.02.02's gevgen_atmo and it used up all my ram memory (see image attached).

Screenshot from 2023-09-20 17-57-00

Is it possible that there is a memory leak? If so, is it an issue with the 3.02.02 version or I is present in other versions?

I run the gevgen_atmo with the following configuration:

FLUX=./kam-nu-20-12-mtn-n3650.d
GEOMETRY=./HyperKamiokande.root
XSECS=./O16_and_H1_xsec.xml
TIME=315360000
ENERGIES=0.1,1.0
LENGTH=m
DENSITY=g_cm3
OUTPUT=HK_atm_bkg
DETECTORPART=Fiducial

gevgen_atmo -f HAKKM:$FLUX[14],$FLUX[12],$FLUX[-14],$FLUX[-12] -r 1 -g $GEOMETRY -t $DETECTORPART -L $LENGTH -D $DENSITY -T $TIME -E $ENERGIES -o $OUTPUT --cross-sections $XSECS --tune G18_10a_02_11a

The table kam-nu-20-12-mtn-n3650.d was obtained from Honda's website (SK with montain max solar cicle) and I obtained O16_and_H1_xsec.xml from the pre-computed xsec using the provided reduce_gxspl.awk file.

I constructed HyperKamiokande.root using a simple root script, which I can share if needed.

pedrosimpas commented

Hi!

A few updates on this issue.

I made some tests and I was able to reduce the options of were the leak is happening.

I tested version 3.02.00 and version 3.04.00. They all have the same memory leak. I also tested to see if the problem is only in
gevgen_atmo or gevgen also has it. And yes, both have it.

Finally, I decided to test different xsection tunes. I found out that if I use the G18_02a_00_000 tune, there is no leak. So it seems the leak is in some function used to calculate the G18_10a_02_11a tune.

I reduced the command that the leak can be reproduced,

gevgen -n 100000 -p 12 -t 1000080160 -e 1.0 --cross-sections O16_xsec_G18_10a_02_11a.xml --tune G18_10a_02_11a

Some information that might be relevant and I didn't include in the previous message.

  • The number of events that uses up all my memory is around 50k (which uses around 15 Gb of ram).

  • Requirements' versions:

log4cpp 1.1.4
LHAPDF 6.5.4
Pythia 6.428
Root 6.28/06

If you need more information, please let me know,

best,

candreop commented

Thanks Pedro. That's useful diagnostic information. We will try to follow up on this.