LHC@home

What is the difference between the following 2 parameters in the app_version section of app_config.xml file, especially as it pertains to multithread apps (LHC ATLAS, MilkyWay N-Body Simulation)?

<app_version>
   <avg_ncpus>x</avg_ncpus>
   <cmdline>--nthreads x</cmdline>
</app_version>

See the BOINC manual:
https://boinc.berkeley.edu/wiki/Client_configuration#Project-level_configuration

avg_ncpus
the number of CPU instances (possibly fractional) used by the app version.
This is used by the BOINC client to calculate various values (work fetch, computer usage, ...)

cmdline
command-line parameters to pass to the program.
The program needs to understand this.
Program can be a scientific app or a wrapper, e.g. vboxwrapper.

I'm familiar with that page and have read it before but it doesn't clarify things much. I understand the difference between cmdline and avg_ncpus in general. I'm specifically wondering about <cmdline>--nthreads x</cmdline> not just cmdline in general. --nthreads in cmdline and avg_ncpus seems to be specifying the same thing and thus seem redundant. However, I've seen people post their app_config files with both entries and I couldn't see why. Would you ever use both in the same app_config and if so under what circumstances? Also, how does one know what cmdline parameters a given program understands?

I understand the difference between cmdline and avg_ncpus

Sorry to say that:
It looks like you don't understand the difference.

As explained in the BOINC manual cmdline passes parameters to the program that is called.
BOINC does not care about what you pass.
It is up to the developers of that program how/if those parameters are understood.

how does one know what cmdline parameters a given program understands?

Under normal circumstances you would call "program {--help|-h|/h"}.
You could also look into the program's manual.
Example:
vboxwrapper's command line parameters can be seen here (usage section):
https://github.com/BOINC/boinc/blob/master/samples/vboxwrapper/vboxwrapper.cpp


Again:
In case of app_config.xml
avg_ncpus controls the BOINC client and not the program that is called.
cmdline controls the program that is called and not the BOINC client.


In case of vboxwrapper a missing cmdline entry automatically results in default values that are passed by the BOINC client.
Those can be taken from app_config.xml or from a couple of other configuration sources.
It's up to the BOINC project developers to set useful values there.


If you set different values for avg_ncpus and nthreads this may lead to unwanted side-effects, typically overloaded systems, work fetch trouble and wrong credit calculation.

Thank you for explaining some more. It seems my suspicion was right that you wouldn't use avg_ncpus and --nthreads command line parameter for the same app. It'll either be redundant or detrimental (if different values are used). avg_ncpus would be the way to control thread usage of multithread apps. I was curious and was able to find a list of command line paramaters for MilkyWay N-Body Simulation but couldn't find them for LHC ATLAS (native). Could you provide a link? Thank you.

Can't guarantee that this is perfect, but it works for me and doesn't throw any errors in the Event Log.

<app_config>
  <app>
      <name>ATLAS</name>
          <cpu_usage>5</cpu_usage>
  </app>
  <app_version>
      <app_name>ATLAS</app_name>
        <plan_class>native_mt</plan_class>
           <avg_ncpus>5</avg_ncpus>
           <cmdline>--nthreads 5</cmdline>
  </app_version>
</app_config>

Could you provide a link?

ATLAS native calls a wrapper script as main program.
The current version of that script is ATLAS_run_atlas_2.87_x86_64-pc-linux-gnu.
It can be found in your projects directory.

  <app>
      <name>ATLAS</name>
          <cpu_usage>5</cpu_usage>
  </app>

This part of your app_config.xml will be ignored by the BOINC client.
It should be removed.

The manual states:
cpu_usage The number of CPU instances (possibly fractional) used by GPU versions of this app.

ATLAS doesn't have a GPU version.
In addition, if you use it for an app that does <cpu_usage> must be enclosed by <gpu_versions> ... </gpu_versions>.

captainjack, yes, I'd expect your app_config to work. There seem to be some redundant/unnecessary entries though. Since you're only trying to modify the native ATLAS version you'd only need the app_version portion of the app_config. In addition, <cmdline>--nthreads 5</cmdline> is redundant since you're already specifying that you want to use 5 CPUs via avg_ncpus. I also believe that the app section is ignored since it's incomplete. Try the following. It's cleaner, shorter and so if you want to change things you're less likely to make an accidental mistake.

<app_config>
    <app_version>
      <app_name>ATLAS</app_name>
      <plan_class>native_mt</plan_class>
      <avg_ncpus>5</avg_ncpus>
    </app_version>
</app_config>

I checked that file and it seems like the only argument that's available is --nthreads. I thought that some time ago when I was trying to figure out how to set up native ATLAS I saw the usage of --memory_size parameter in the forums. I added it to the app_config file (can't remember what made me think I needed it) but it seems like it's invalid and is just ignored. I'll have to delete it then. Was it ever used in the past?

--memory_size parameter ... Was it ever used in the past?

I don't think it has ever been used for ATLAS native.
It is mentioned very often due to the fact that it is used by vboxwrapper for years.
Most people just copy/paste those options without thinking about whether it makes sense.

I use the --memory_size_mb one, since I force atlas to use 1 WU and unlimited on the web settings this tells BOINC each WU use 10 GB, since BOINC thinks its using 10 GB anyway I just let it go with 10 GB.

It seems like you're using the VBox version of ATLAS not native. According to a post above, that argument is valid in the VBox version not native though.