LHC@home

On computers with lots of cores it might be worth to set up additional BOINC client instances.

This is more work than I'm willing to do.
I have a much better idea. Add BOINC commands that tell the server what to do:

<app_config>
<app>
    <name>ATLAS</name>
    <!-- Xeon E5-2699 v4  22c44t  32 GB RAM L3 Cache = 55 MB  -->
    <maintain>18</maintain>
    <max_concurrent>16</max_concurrent>
</app>
<app_version>
    <app_name>ATLAS</app_name>
    <plan_class>native_mt</plan_class>
    <avg_ncpus>1</avg_ncpus>
    <cmdline>--nthreads 1</cmdline>
</app_version>
<app>
    <name>sixtrack</name>
    <maintain>9</maintain>
    <max_concurrent>6</max_concurrent>
</app>
<app>
    <name>Theory</name>
    <maintain>44</maintain>
</app>
<app>
    <name>CMS</name>
    <maintain>0</maintain>
</app>
</app_config>

And even better would be:

<app_config>
<app>
    <name>ATLAS</name>
    <!-- Xeon E5-2699 v4  22c44t  32 GB RAM L3 Cache = 55 MB  -->
    <priority>1</priority>
    <max_concurrent>16</max_concurrent>
</app>
<app_version>
    <app_name>ATLAS</app_name>
    <plan_class>native_mt</plan_class>
    <avg_ncpus>1</avg_ncpus>
    <cmdline>--nthreads 1</cmdline>
</app_version>
<app>
    <name>sixtrack</name>
    <priority>3</priority>
</app>
<app>
    <name>Theory</name>
    <priority>2</priority>
</app>
<app>
    <name>CMS</name>
    <priority>0</priority>
</app>
</app_config>

The local BOINC client will simply ignore xml tags that are not defined for app_config.xml.

Among those ignored tags are:
<maintain>18</maintain>
<priority>1</priority>
<whatever_undefined_tag_is_written_here>foobar</whatever_undefined_tag_is_written_here>

Allowed tags are listed here:
https://boinc.berkeley.edu/wiki/Client_configuration

<app_config>
   [<app>
      <name>Application_Name</name>
      <max_concurrent>1</max_concurrent>
      [<report_results_immediately/>]
      [<fraction_done_exact/>]
      <gpu_versions>
          <gpu_usage>.5</gpu_usage>
          <cpu_usage>.4</cpu_usage>
      </gpu_versions>
    </app>]
   ...
   [<app_version>
       <app_name>Application_Name</app_name>
       [<plan_class>mt</plan_class>]
       [<avg_ncpus>x</avg_ncpus>]
       [<ngpus>x</ngpus>]
       [<cmdline>--nthreads 7</cmdline>]
   </app_version>]
   ...
   [<project_max_concurrent>N</project_max_concurrent>]
   [<report_results_immediately/>]
</app_config>

The local BOINC client will simply ignore xml tags that are not defined for app_config.xml.
Among those ignored tags are:
<maintain>18</maintain>
<priority>1</priority>

Duh. So invent them.

The local BOINC client will simply ignore xml tags that are not defined for app_config.xml.
Among those ignored tags are:
<maintain>18</maintain>
<priority>1</priority>

Duh. So invent them.

This would be a BOINC client change, independent from LHC@home.
Feel free to post your request here:
https://github.com/BOINC/boinc

The local BOINC client will simply ignore xml tags that are not defined for app_config.xml.
Among those ignored tags are:
<maintain>18</maintain>
<priority>1</priority>

Duh. So invent them.

It is an open source project. If you would like to invent them you are able to do so. It is much less work to learn and use new management tools though. Like I said before, it is easy to use LXD once you learn.

I do not intend to be rude but I have to say that you are in the top 1 percent of contributors and I know that must be a work of it's own. Why not work a little harder for a short period of time so you will not have to work so hard for years to come? Why must the software developers appease your desire not to work when they also do not desire to work?

I thought nT was in production but it's limited to a fearful 10 WUs per rig. Because of the RAM-hungry ATLAS WUs I have to run ST to fill out my threads. I think BOINC runs best with fewer projects but stuck with running three now.

How many per rig should we increase it to?

Project-Server for native-Theory 1.1 is now empty, but
this Task is shown from June and not deleted:
https://lhcathome.cern.ch/lhcathome/workunit.php?wuid=116083221

How many per rig should we increase it to?

My suggestion would be to calculate it based on <p_ncpus> which is already included in the client's scheduler request.

Either:
max_tasks_to send=p_ncpus

Or:
max_tasks_to send=p_ncpus+8 (allows a small buffer)

In addition the max number of tasks to be send per request should also be limited (see other projects like Milkyway).
This would avoid that a single request from a high potential host clears the server's buffer.
Suggestion: 8

My suggestion would be to calculate it based on which is already included in the client's scheduler request.

It was already limited to 2 tasks per p_npcpus but also had a maximum limit of 10. I have removed that maximum limit.