LHC@home

Start reading before writing :)

Theory and ATLAS are limited server side to avoid swamping the clients with too many tasks.
ATLAS uses 4 * #cores, Theory ... (don't know).


On your 12 core computer you can use "3" #cores on the web prefs page to allow 12 ATLAS tasks to be downloaded.
Then use this app_config.xml:

<app_config>
   <app_version>
       <app_name>ATLAS</app_name>
       <plan_class>vbox64_mt_mcore_atlas</plan_class>
       <avg_ncpus>1.0</avg_ncpus>
       <cmdline>--memory_size_mb 3900</cmdline>
   </app_version>
</app_config>

Be aware:
ATLAS uses a 2nd memory configuration parameter that can't be modified via app_config.xml.
This limits the number of ATLAS tasks on your computer to 11 (may be less).

Theory tasks usually start with #events = 100000.
In very rare cases they don't finish within the 10 day limit.

If the long runtime is
- not caused by a local issue and
- mcplots does not get enough valid results for a given set of input parameters
the same task type is reissued with a lower #events.

This reduction may happen repeatedly until enough valid results are returned.
A statement like "Cern-IT have no interest" is simply wrong.

Why do you ask anybody else?

?
No Multicore (Atlas or Theory) in Windows atm.

If the long runtime is
- not caused by a local issue and
- mcplots does not get enough valid results for a given set of input parameters
the same task type is reissued with a lower #events.

10 days limit is not enough for this task:
https://lhcathome.cern.ch/lhcathome/workunit.php?wuid=221471152

It is running 2 days already (24/7) and have done only 10900 events (of 100000 total).

Previous attempt on PC 10836791 confirm the problem.

I was not aware that obviously ALL Theory tasks are faulty now.
I downloaded them on several hosts, and all tasks errored out after some time, up to after more than 1 hour :-(

So I am wondering why no-one back at LHC&Home stops the download of Theory tasks and empties the queue. The result is a real waste of ressources on the volunteers' side.

I'm draining the job queue in anticipation of a WMAgent upgrade tomorrow. Jobs will probably start to run out around 1200GMT, so set No New Tasks late tonight or early in the morning.

The update is done, and I've injected a new workflow. With any luck new tasks/jobs will be available within the next hour or so.

...and... we're off again! Thanks everyone.

Your recent logfiles report this lines:

2021-09-22 04:23:39 (18124): Setting Memory Size for VM. (10200MB)
2021-09-22 04:23:39 (18124): Setting CPU Count for VM. (8)

This happens if you set #cores at the web prefs to "8" or "unlimited".
The RAM size allocated for the VM is then calculated to 3000MB + 8 * 900MB = 10200MB.
The same RAM size is used by the BOINC client to calculate the total RAM requirement per ATLAS task even if you tweak the local setting via app_config.xml

Limited by the #CPUs (12) your BOINC client would run just 1 ATLAS task concurrently using 8 cores and 10.2GB RAM.



If you set the web prefs to #core=1 the RAM requirements would be calculated to 3000MB + 1 * 900MB = 3900MB.
This would allow your BOINC client to run 12 ATLAS VMs concurrently using 46.8GB RAM, but as you already noticed the server would send only 4 tasks.


Setting the web prefs to #cores=2 would allow 8 tasks to be send.
Your computer could run 6 of them concurrently (total CPU limit) and each task would allocate 4.8GB => 28.8GB in total

Same with #cores=3
max #tasks sent: 12
running concurrently: 4 (total CPU limit)
total RAM used: 22.8GB

This setting can be tweaked by an app_config.xml as already explained.
If <avg_ncpus>1.0</avg_ncpus> and <cmdline>--memory_size_mb 3900</cmdline> are set this would correspond to a 1-core setup but the server would still send 12 tasks.
Nonetheless your BOINC client would use a value of 5.7GB to estimate whether there's enough free RAM to start an additional task.


To make all of that work you need to
- reload the BOINC config files (or restart the BOINC client) after you modified app_config.xml.
- ensure all tasks currently in your work buffer are created with the same prefs settings.

- ensure all tasks currently in your work buffer are created with the same prefs settings.

Theory and ATLAS are limited server side to avoid swamping the clients with too many tasks.
ATLAS uses 4 * #cores, Theory ... (don't know).


On your 12 core computer you can use "3" #cores on the web prefs page to allow 12 ATLAS tasks to be downloaded.
Then use this app_config.xml:

<app_config>
   <app_version>
       <app_name>ATLAS</app_name>
       <plan_class>vbox64_mt_mcore_atlas</plan_class>
       <avg_ncpus>1.0</avg_ncpus>
       <cmdline>--memory_size_mb 3900</cmdline>
   </app_version>
</app_config>

Be aware:
ATLAS uses a 2nd memory configuration parameter that can't be modified via app_config.xml.
This limits the number of ATLAS tasks on your computer to 11 (may be less).

David Cameron wrote this:
By default this is 1

You may try the following setup.

The web preferences for this computer should be set to 3 CPUs.

Then use an app_config.xml to reduce ATLAS to 1 CPU.
See section "Project-level configuration"

Slightly OT but worth to be mentioned:
Since you attached roughly 100 cores to LHC@home it is recommended that you use a local HTTP proxy.
https://lhcathome.cern.ch/lhcathome/forum_thread.php?id=5473