Problem 2178

Summary: Multi-thread with physics models
Product: Geant4 Reporter: Jaeyoung Jeong <jyjeong59>
Component: processes/hadronicAssignee: dennis.herbert.wright
Status: RESOLVED WONTFIX    
Severity: normal CC: Alberto.Ribon, Vladimir.Ivantchenko
Priority: P4    
Version: 10.5   
Hardware: PC   
OS: Linux   

Description Jaeyoung Jeong 2019-07-22 11:37:34 CEST 
Hello, my name is Jaeyoung Jeong in Korea.

When I simulated exampleB1 in the QGSP_INCLXX model with more than 3 CPUs and 1e9 primary particles, it didn't use the whole CPU performance and the %CPU decreased slowly.

When I used QGSP_BERT, it works well with 300% %CPU. I think the multi-thread has something wrong,,

The same problem has occurred with different input files.

Best regards,
Jaeyoung Jeong
Comment 1 Ivana Hrivnacova 2019-07-22 11:55:05 CEST 
Hello,

As the problem happens when the QGSP_INCLXX model is included, then it should be related to Geant4 hadronic processes and not to the basic example.
I have modified the Product/Component to Geant4/processes/hadronic.

Best regards,
Comment 2 dennis.herbert.wright 2019-07-22 19:59:26 CEST 
It sounds like the multithreading is OK, but something is wrong with INCLXX, perhaps in the way it handles multithreading.  I will check with the INCLXX author.
Comment 3 dennis.herbert.wright 2019-08-09 23:07:04 CEST 
Some work has been done by the INCLXX developers and some fixes of multithreading were completed for the Geant4 10.5 patch 01 release in April.  I'm not sure these fixed your problem, but please try patch 01 if you have not already done so.
Comment 4 Vladimir.Ivantchenko 2019-08-17 19:25:51 CEST 
Hello,

during Geant4 testing we run different Physics Lists with high statistics including tests in MT mode. Of course, we cannot find all rare problems. In order to study rare problems we need very precise description of the use case.

VI
Comment 5 Jaeyoung Jeong 2019-08-17 19:53:21 CEST 
(In reply to Vladimir.Ivantchenko from comment #4)
> Hello,
> 
> during Geant4 testing we run different Physics Lists with high statistics
> including tests in MT mode. Of course, we cannot find all rare problems. In
> order to study rare problems we need very precise description of the use
> case.
> 
> VI

Hello, thank you for your reply.


I studied about the muon production at the reaction between high energy protons and graphite.


Basically, I changed Detectorconstruction and Steppingaction in the exampleB1 in Geant4.10.5. The target was 1.80 g/cm3 graphite and produced muons were counted in steppingaction. I think I can't describe my input file well, so, If you don't mind, I will send you input files.

I used Geant4 version 4.10.5.p01.


Additionally, I had a Multithreading problem with just exampleB1 with the other physics models(QGSP_BIC, QGSP_INCLXX).

Best regards,
Jaeyoung Jeong
Comment 6 Jaeyoung Jeong 2019-08-17 19:57:02 CEST 
(In reply to dwright from comment #3)
> Some work has been done by the INCLXX developers and some fixes of
> multithreading were completed for the Geant4 10.5 patch 01 release in April.
> I'm not sure these fixed your problem, but please try patch 01 if you have
> not already done so.

Thank you for your reply,

I already updated the version to 4.10.5.p01.

I will try other solutions. Thank you for your advise again.

Best regards,
Jaeyoung Jeong.
Comment 7 Alberto.Ribon 2019-12-13 11:55:47 CET 
Given that we are unable to reproduce the problem in our tests, we will not be able to study and fix it, so we close it.