Development for LHC@home

There's something wrong regarding the runtime calculation.

Even when running single-core tasks, the cpu-time is ever (mostly) higher than the elapsed time.
When you have idle cores the application is stealing from the free core(s).

See my Linux tasks and my Windows Vbox tasks.

Right, this explains the higher CPU-times.
In the mentioned examples I don't understand why the runtime is much higher than the difference between app start and app finish.

BTW:
There is no visible delay at app start or app finish.

Have 7 tasks in parallel, but slot-Nr. are shown up to 21!
Maybe, they are not deleted after finishing?

Yes there was an issue with a previous image where the slot directories were not be clean. Let me know if that is still the case.

Have cleaned the Slot-Nrs and will control it. Thank you.

A question about the length of the tasks.
They have a duration for example Pythia8 from 4 or 5 hours.
Are this tasks the same as for production?
Thinking they need longer. Can this be?

There are tasks that run for a couple of hundred seconds and some for a couple of hours. Its probably the normal fluctuations (different job types, different codes, different events types, ...). Using the old vbox app there have also been jobs that run for a couple of hours. Would be weird if the native jobs run slower compared to running within a VM.

Have cleaned the Slot-Nrs and will control it. Thank you.

A question about the length of the tasks.
They have a duration for example Pythia8 from 4 or 5 hours.
Are this tasks the same as for production?
Thinking they need longer. Can this be?

The same jobs are run as in production. There is a mixture of job types and parameters so it is normal to have different runtimes.

It's not against the very good work from Laurence for this new concept.
Sorry, if it is to criticle.

Thanks, negative feed back is important. It is better to have it here and try to resolve the issue than in production.

Have 1 failed task in a row of valids.
Reason unknown.
https://lhcathomedev.cern.ch/lhcathome-dev/result.php?resultid=2755317

Same here after 8.5 hours run-time: EXIT_CHILD_FAILED

09:33:23 2019-02-26: cranky-0.0.24: [INFO] Running Container 'runc'.
18:11:23 2019-02-26: cranky-0.0.24: [ERROR] Container 'runc' failed.

https://lhcathomedev.cern.ch/lhcathome-dev/result.php?resultid=2755265

had one of those to two days ago:
195 (0x000000C3) EXIT_CHILD_FAILED
https://lhcathomedev.cern.ch/lhcathome-dev/result.php?resultid=2754240
Dont think that this has something to do with the problem, but anyway:
I did a router restart while the task was running, so temporary there was no internet access available for the task.

There's something wrong regarding the runtime calculation.

Even when running single-core tasks, the cpu-time is ever (mostly) higher than the elapsed time.
When you have idle cores the application is stealing from the free core(s).

See my Linux tasks and my Windows Vbox tasks.

Right, this explains the higher CPU-times.
In the mentioned examples I don't understand why the runtime is much higher than the difference between app start and app finish.

BTW:
There is no visible delay at app start or app finish.

An upgrade of my BOINC client may have solved that issue.

... An upgrade of my BOINC client may have solved that issue.

Well, my fault.
I checked the first tasks after a reboot when the host was not yet under full load.
Under full load the runtime calculation still differs significantly compared to "finish time - starting time".
See examples:
https://lhcathomedev.cern.ch/lhcathome-dev/result.php?resultid=2755583
https://lhcathomedev.cern.ch/lhcathome-dev/result.php?resultid=2755593

Mostly the CPU-times are (much) higher than the elapsed times.

Run time 2 hours 10 min 54 sec
CPU time 3 hours 3 min 55 sec

https://lhcathomedev.cern.ch/lhcathome-dev/result.php?resultid=2755539

Run time 1 hours 12 min 56 sec
CPU time 1 hours 40 min 16 sec

https://lhcathomedev.cern.ch/lhcathome-dev/result.php?resultid=2755545

Your example:
https://lhcathomedev.cern.ch/lhcathome-dev/result.php?resultid=2755539

The log shows 2h 10m 38s between "wrapper starting" and "called boinc_finish".
Runtime shows 2h 10m 54s.

A difference of 16s.
Not very much, but in my eyes too much on a modern computer to fork/shutdown a process.

Was your computer fully loaded?
Is it bare metal or a VM?



My full load example on a bare metal computer shows a time difference of more than 10 minutes (!).
This seems to be stupid.
https://lhcathomedev.cern.ch/lhcathome-dev/result.php?resultid=2755583

Your example:
https://lhcathomedev.cern.ch/lhcathome-dev/result.php?resultid=2755539
.
.
Was your computer fully loaded?
Is it bare metal or a VM?

That task was on a 4-core Linux VM on a Windows host.
Only 2 cores of the VM were used.

In your task example the run-time must be wrong. It must be a BOINC/vboxwrapper issue.
The cpu-time looks like the same as reported on the second last line in the result.
The elapsed and cpu time being exactly the same is suspicious.

SL 76 show under TOP:
2 GByte RAM of Swap in use with 0 free.
The normal Memory is only for 50% in use (10Gbyte of 20 GByte).
Is this also in other native Linux so?

Is this also in other native Linux so?

Ubuntu 18.10

top - 12:41:31 up  3:31,  1 user,  load average: 1,18, 1,29, 1,34
Tasks: 218 total,   2 running, 216 sleeping,   0 stopped,   0 zombie
%Cpu(s):  0,3 us,  0,7 sy, 28,2 ni, 70,8 id,  0,0 wa,  0,0 hi,  0,0 si,  0,0 st
MiB Mem :   5960,3 total,   3671,1 free,    890,1 used,   1399,1 buff/cache
MiB Swap:   1186,4 total,   1186,4 free,      0,0 used.   4771,1 avail Mem 

  PID USER      PR  NI    VIRT    RES    SHR S  %CPU  %MEM     TIME+ COMMAND                 
23512 boinc     39  19  476848  89408  42576 S  72,1   1,5  58:46.48 Herwig                  
23399 boinc     39  19  292024  19660  13056 R  43,2   0,3  35:53.56 rivetvm.exe             
 7573 boinc     39  19    4132    184    144 S   0,3   0,0   0:00.01 sleep                   
 1370 boinc     30  10  166892  16492  13028 S   0,0   0,3   0:26.20 boinc                   
 7521 boinc     39  19    4132     36      0 S   0,0   0,0   0:00.00 sleep                   
21989 boinc     30  10    6052   3312   2832 S   0,0   0,1   0:06.79 wrapper_26015_x         
21991 boinc     39  19   20256   3600   3216 S   0,0   0,1   0:00.01 cranky-0.0.24           
22769 boinc     39  19  609124   6624   2024 S   0,0   0,1   0:00.04 runc                    
22779 boinc     39  19   17728    204      0 S   0,0   0,0   0:00.01 job                     
22796 boinc     39  19   18664   1748    628 S   0,0   0,0   0:02.95 runRivet.sh             
23398 boinc     39  19   18256    796      0 S   0,0   0,0   0:00.04 rungen.sh               
23400 boinc     39  19   18796   1888    624 S   0,0   0,0   0:03.63 runRivet.sh

Hi Crystal,
the same for you!
Theory-native use the Swap-Memory instead of the free normal Memory.
Atlas-native does this not!
Swap-Memory is normaly a Disk-File.

Hi Crystal,
the same for you!
Theory-native use the Swap-Memory instead of the free normal Memory.
Atlas-native does this not!
Swap-Memory is normaly a Disk-File.

MiB Swap:   1186,4 total,   1186,4 free,      0,0 used

This means:
Swap total: 1186,4 MB
Swap free: 1186,4 MB
Swap used: 0,0 MB

ATM no swap is used on this host.

Thank you, sorry,
SL69 total used free
SL76 total free used.
Have not seen, it is changed in TOP.

In the -dev Folder of Boinc is a cranky-0.0.24 and a cranky-0.0.25.
The tasks show the protocol with cranky-0.0.24.
08:50:27 (20616): wrapper: running ../../projects/lhcathomedev.cern.ch_lhcathome-dev/cranky-0.0.24 ()