LHC@home

Bonjour.
J'aimerais calculer 3 unités en meme temps mais 1 seule par sous projet.
C'est a dire une unité atlas native avec 12 threads,une unité cms virtualbox avec 4 threads et une unité theory virtualbox avec un thread.
J'ai utilisé Claude AI et https://statseb.boinc-af.org/app_config_generator.py pour générer le fichier.
Quelqu'un pourrait il le vérifier et éventuellement l'optimiser.
Je vous remercie.

Hello.
I would like to calculate 3 units at the same time but only 1 per sub-project.
That is to say a native atlas unit with 12 threads, a virtual cms box unit with 4 threads and a virtual theory box unit with one thread.
I used Claude AI and https://statseb.boinc-af.org/app_config_generator.py to generate the file.
Could someone check it and possibly optimize it.
Thank you.

<app_config>
<app>
<name>ATLAS</name>
<max_concurrent>1</max_concurrent>
<gpu_versions>
<gpu_usage>1.000000</gpu_usage>
<cpu_usage>12.000000</cpu_usage>
</gpu_versions>
<fraction_done_exact>1</fraction_done_exact>
<report_results_immediately>1</report_results_immediately>
</app>
<app>
<name>CMS</name>
<max_concurrent>1</max_concurrent>
<gpu_versions>
<gpu_usage>1.000000</gpu_usage>
<cpu_usage>4.000000</cpu_usage>
</gpu_versions>
<fraction_done_exact>1</fraction_done_exact>
<report_results_immediately>1</report_results_immediately>
</app>
<app>
<name>Theory</name>
<max_concurrent>1</max_concurrent>
<gpu_versions>
<gpu_usage>1.000000</gpu_usage>
<cpu_usage>0.000000</cpu_usage>
</gpu_versions>
<fraction_done_exact>1</fraction_done_exact>
<report_results_immediately>1</report_results_immediately>
</app>
<app_version>
<app_name>ATLAS</app_name>
<plan_class>native_mt</plan_class>
<cmdline></cmdline>
<avg_ncpus>12.000000</avg_ncpus>
<ngpus>1.000000</ngpus>
</app_version>
<app_version>
<app_name>CMS</app_name>
<plan_class>vbox64_mt_mcore_cms</plan_class>
<cmdline></cmdline>
<avg_ncpus>4.000000</avg_ncpus>
<ngpus>1.000000</ngpus>
</app_version>
<app_version>
<app_name>Theory</app_name>
<plan_class>vbox64_theory</plan_class>
<cmdline></cmdline>
<avg_ncpus>1.000000</avg_ncpus>
<ngpus>1.000000</ngpus>
</app_version>
<project_max_concurrent>3</project_max_concurrent>
<report_results_immediately>1</report_results_immediately>
</app_config>

---

A 12-core setup is a rather bad idea for ATLAS.
Each task processes 400 events.
Spread over 12 cores this will leave 4 working cores and 8 idle cores at the end of each task (averaged).

More efficient would be to use an 8-core setup (0 idle cores).
If you don't mind the longer total walltime you may even use a 4-core setup as this is known to be the most efficient one.


As for the app_config.xml

Your AI suggested one is not really good.
It would have been better to follow the rules from the BOINC manual:
https://boinc.berkeley.edu/wiki/Client_configuration#Project-level_configuration
- don't set optional tags that are not required
- don't set tags that are unused here, e.g. GPU options for CPU apps
- don't set tags to repeat values that are already set in client_state.xml, e.g. 4 cores for CMS
- resort for better readability (keep app and app_version together)

<app_config>
  <app>
    <name>ATLAS</name>
    <max_concurrent>1</max_concurrent>
  </app>
  <app_version>
    <app_name>ATLAS</app_name>
    <avg_ncpus>4.0</avg_ncpus>
  </app_version>
  <app>
    <name>CMS</name>
    <max_concurrent>1</max_concurrent>
  </app>
  <app>
    <name>Theory</name>
    <max_concurrent>1</max_concurrent>
  </app>
  <report_results_immediately>1</report_results_immediately>
</app_config>

There are no GPU versions listed at https://lhcathome.cern.ch/lhcathome/apps.php so I think the <gpu_versions> sections are not useful and should be removed for simplicity and readability.
<max_concurrent> and <project_max_concurrent> should work together correctly as specified.
Specifying <ngpus>1.000000</ngpus> may have undesirable side effects if there is only one GPU in the system.
Is there a specific reason for the empty <cmdline> elements?
<report_results_immediately> may be useful in some circumstances but I would use the default unless there is a specific reason to specify it.

Indenting nested elements improves readability, even if the system doesn't care, e.g.,

<app>
    <name>ATLAS</name>
    <max_concurrent>1</max_concurrent>
    <gpu_versions>
        <gpu_usage>1.000000</gpu_usage>
        <cpu_usage>12.000000</cpu_usage>
    </gpu_versions>
    <fraction_done_exact>1</fraction_done_exact>
    <report_results_immediately>1</report_results_immediately>
</app>

the Code button will let you insert text while preserving leading white space in Forum posts.

---