Development for LHC@home

This version provides the updated docker wrapper v18.

https://lhcathomedev.cern.ch/lhcathome-dev/workunit.php?wuid=2709704
First docker Task in -dev running.
Same problem as in Production, Squid is not running.
Squid is running as a VM in Virtualbox.
Have no idea, how to get a solution.

In reply to maeax's message of 18 Feb 2026:

https://lhcathomedev.cern.ch/lhcathome-dev/workunit.php?wuid=2709704
First docker Task in -dev running.

Be prepared. that this "pp jets 13000 210 - pythia8 8.244 CP1-CR1" seems to be a long running job.
From 7 attemps: 2 success. 1 failure and 4 unknown.

Thank you for the Info,
10 day max or more, we will see.
btw: You wrote yesterday, Docker Tasks are faster ;-)

In reply to maeax's message of 18 Feb 2026:

btw: You wrote yesterday, Docker Tasks are faster ;-)

Compared to VirtualBox on a Windows host.

You can only check this by running exactly the same Theory "job run" with the same parameters on both platforms under comparable load.
I was very surprised that the difference was so big, because with VirtualBox we see elapsed time and cpu-time always close together and
wirh WSL container the elapsed time of a task is much longer than the measured cpu-time, what is strange. When looking at the cpu-usage of the Theory containers you see very often cpu-usage over 100%.
Example: pythia8.exe and the belonging rivetvm.exe together 103%.
I suppose the used cpu-time is badly measured by BOINC. The main process on the Windows host is vmmem and when running one single task, this process on a quad core (2 cores/4 threads) jumps between 25% to 40%.
I've limited WSL to max 3 cores and 3 GB RAM. When running 3 tasks the CPU usage of vmmem is between 60% and 75%. But at the end the elapsed times are much shorter. See my last example:

pp zinclusive 8000 -,-,100,170 - pythia8 8.306 tune-AU2m 100000 650

vbox:   Run time  5 hours  7 min 54 sec        CPU time	4 hours 57 min 50 sec
docker: Run time  1 hours 54 min 10 sec        CPU time 1 hours 40 min 49 sec

There is one big disadvantage at the moment of the use of docker for Theory tasks: When you reboot the host system, all work done before is lost and the task restarts from scratch.

theory_2922-4811546-642_0
Laufzeit 5 Tage 8 Stunden 58 min. 56 sek.
CPU Zeit 5 Tage 3 Stunden 33 min. 54 sek.
Priority 0
Prüfungsstatus Gültig
Punkte 8,112.07
max. FLOPS des Gerätes 7.55 GFLOPS
Anwendungsversion Theory Simulation v302.10 (docker)
windows_x86_64

One more Docker-Task now with 6 Days atm is running also.
https://lhcathomedev.cern.ch/lhcathome-dev/workunit.php?wuid=2709704

Name Theory_2922-4797518-490
Anwendung Theory Simulation
batch 524986
erstellt 5 Feb 2026, 3:20:53 UTC
autorisiertes Ergebnis 3700295
gewährte Punkte 7,688.01

next Task started.
Squid is active, have no way to control squid in Theory task.
Squid working inside of a VM.

I killed some neverending wus
Running Time 1 day 18 hours 13 minutes 12 seconds
CPU Time 1 day 0 hours 43 minutes 58 seconds

Well I got to run a few (43) but we are out of work with this version but have 2 still running and plenty over at production with that version 302.10

Extension to https://lhcathomedev.cern.ch/lhcathome-dev/forum_thread.php?id=739&postid=9258

I did some new tests to compare the efficiency of VBox versus Docker (WSL) processing exactly the same job.

A short job: ppbar top-mc 1960 - - herwig++ 2.7.1 default 100000 715

docker: Run time  18 min 35 sec        CPU time	14 min 55 sec
vbox:   Run time  26 min 59 sec        CPU time 26 min  8 sec


A longer job: ppbar jets 1960 65 - pythia8 8.308 tune-monash13 100000 714

docker: Run time  0 hour  51 min 20 sec        CPU time	0 hour  43 min 12 sec
vbox:   Run time  2 hours 25 min 49 sec        CPU time 2 hours 21 min 13 sec

I don't have many to compare to VB but the Dockers are usually fairly fast but have some that are close to 10hrs and some as long as over 2 days to over 4 days
Theory Simulation v7.66 (docker)
windows_x86_64

I found this one
https://lhcathomedev.cern.ch/lhcathome-dev/workunit.php?wuid=2718012

You’re comparing apples to pears.
In that workunit really the same job (pp jets 13000 400 - pythia8 8.315 stringzold-off 100000 691) is processed twice.
Your's with VirtualBox and the other with Docker, but on two totally different hardware (CPU, RAM etc.). 13th Gen Intel(R) Core(TM) i7-13700 versus AMD Ryzen 7 6800H.
The load on the system should also be the same - % threads, % CPU, other activities on the test-system etc.

To compare the performance on both platforms one should run exactly the same sub-job on the same computer under the same conditions.

I have run a few more tests to compare the performance of Docker against VirtualBox.
It is abundantly clear that Theory tasks on Windows WSL (Docker) are processed much faster.
Vbox and docker did exactly the same job on the same hardware and same conditions. The result of the last two comparison tests:

job: pp jets 7000 80,-,1060 - pythia8 8.315 stringzold-on 100000 717

vbox:   Run time  2 hours  7 min 17 sec        CPU time 1 hour  49 min 41 sec
docker: Run time  0 hour  56 min 27 sec        CPU time	0 hour  49 min 16 sec


job: pp w1j 7000 250 - pythia6 6.428 z2-lep 100000 717

vbox:   Run time  2 hours  2 min 12 sec        CPU time 1 hour  44 min 45 sec
docker: Run time  0 hour  44 min 16 sec        CPU time	0 hour  38 min 45 sec

Testing a longer running pythia8 job on the same computer on both platforms:

job: pp jets 7000 150,-,2160 - pythia8 8.313 vincia-default 100000 718

vbox:   Run time  6 hours 48 min 20 sec        CPU time 6 hour  41 min 39 sec
docker: Run time  3 hours 44 min  1 sec        CPU time	3 hours 17 min  5 sec