LHC@home

Some novelty weirdness.
Yesterday morning I looked at my only Linux box, Minty, and saw only a black screen with a flashing cursor in the top left corner. Unresponsive. Had to leave it that way until the evening. At 04:27:23 yesterday, 2 x Sixtracks reported Validate Errors and this Theory-Native also broke so I suspect the weirdness happened around this time.
Several reboots. No joy. Some Googling and the live USB eventually got me in far enough to find the disk completely full, hence the unresponsiveness. Deleted some stuff to allow enough free space for it to boot properly (9.3GB free of 1TB). Left Disk Usage Analyser looking overnight. This morning it had found a HUUUGE Timeshift snapshot within which the cvmfs file had grown to 885GB !!
Left it deleting that and a couple of regular sized snapshots while out at work.

Back home this evening to find 952GB now free 8¬) but further weirdness in that, without me having started Boinc, during the day, it has manged to complete 2 x Theory-Natives (and a couple of faulty ones from -dev) and my Tasks page has a VBox Theory, 2 x Sixtracks and 2 x Theory-Natives in progress.
Some Boinc processes are listed in "top" but it's not "running" on the desktop (taskbar icon not highlighted), it's not set to run-at-start-up, and I didn't start it. When I did start Boinc, all seems well. The VBox task running and another Theory-Native from -dev, and the icon is highlighted now.

All seems back to normal now. Think I'll just do manual Snapshots in case it was Timeshift that had caused the problem.

Tried a manual Timeshift Snapshot just now and cancelled it after about 10 mins when it was estimating a further 5hrs, and rising, to complete. I guess it gets confused if the files are actively changing and it has to continually add to the snapshot so I've disabled automatic snapshot schedule and will only do manual ones when Boinc isn't running. Guess I've just been lucky not to have encountered this problem before.

With Boinc not running, a manual snapshot took <10mins, confirming that Timeshift and Boinc don't play well together.

runRivet] Wed Sep 18 19:32:11 UTC 2019 [boinc pp jets 8000 25 - herwig7 7.1.3 softTune 100000 122]
This Task finished not successful:
https://lhcathome.cern.ch/lhcathome/result.php?resultid=246070380

TheoryN_2279-789462-146_1 works not for three User:
https://lhcathome.cern.ch/lhcathome/workunit.php?wuid=125694976
09:54:35 CET +01:00 2019-10-29: cranky-0.0.29: [INFO] ===> [runRivet] Tue Oct 29 08:54:35 UTC 2019 [boinc pp jets 8000 25 - pythia8 8.235 cr1 77000 146]

TheoryN_2279-789477-148 - [boinc pp jets 8000 25 - herwig++ 2.5.2 LHC-UE-EE-3-7000 100000 148]
https://lhcathome.cern.ch/lhcathome/workunit.php?wuid=126046148

TheoryN_2279-789480-148 - [boinc pp jets 8000 25 - herwig++ 2.6.0 LHC-UE-EE-4-CTEQ6L1 100000 148]
https://lhcathome.cern.ch/lhcathome/workunit.php?wuid=126060226

TheoryN_2279-805270-155 - [boinc pp zinclusive 7000 -,-,50,130 - madgraph5amc 2.6.2.atlas default 100000 155]
https://lhcathome.cern.ch/lhcathome/workunit.php?wuid=126146734

TheoryN_2279-789271-184 - [boinc pp jets 8000 25 - pythia6 6.428 380 100000 184]
https://lhcathome.cern.ch/lhcathome/workunit.php?wuid=126426252

Theory_2279-789369-186 - [boinc pp jets 8000 25 - pythia8 8.212 tune-AU2lox 100000 186]
https://lhcathome.cern.ch/lhcathome/workunit.php?wuid=126721335

Theory_2279-789417-188 - [boinc pp jets 8000 25 - pythia8 8.230 tune-4cx 100000 188]
https://lhcathome.cern.ch/lhcathome/workunit.php?wuid=127248262

I'm installing cvmfs on my machines. I don't know where I can do this question and I didn't find answers, so here it is.

I have got an i5, so 4 threads. I set 4 native tasks at time. Why are there so many processes running concurrently using 30-50% instead of only 4 processes using 100%? Is this efficient?

Is this efficient?

Short answer:
Yes


Long answer:
The "program" is divided in different parts each of them represents a specialized function or program step.
Program parts are chained via pipes and/or buffers and each function/step writes it's results into this pipes/buffers where it can be read by the next step.
A full buffer causes a function to sleep and give the next step the chance to catch up.

What you see in your diagnostic tool is an average cpu usage, e.g. for the last 3 seconds.
30% process A, 70% process B would mean:
Process A used 0.9 s to fill the output buffer and process B used 2.1 s to work on that results.
The total work percentage of that app would be 100%.

Thank you very much computezrmle. I appreciated both answers and their order.

I have been suspecting there was a logic behind but I don't know about CPUs so much.

Now that maeax has the last nT WU running will nT 1.1 WUs be released to the public???

Sorry Aurum, don't have the last native Theory 1.1.
When this last task is finished, native Theory 1.1 can be removed from the Production-Server, that's correct.

But maeax appears to be the world record holder for longest running nTheory WU :-)
Point is there's only one left as shown on Server Stats. Are you saying nT 1.1 is done?
Just wondering if we'll get more nT WUs.

Theory Native migrate to application 300.02 (native_theory) so same but under another version on project side. To get new/more this would need to used in your settings.

Thanks Gunde, I checked the Preferences/Theory Simulation box and 300s started flowing down.

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.

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

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.

Since you have your computers hidden in your profile I do not know if this will apply to you but:

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

It's really easy to do this on a Linux system that can run LXD. You can limit the ram and/or CPU threads for each container manually or by using profiles. Even though the following information is a few years old it is still good. It is written by the lead developer of LXD:

https://stgraber.org/2016/03/11/lxd-2-0-blog-post-series-012/

I hope that is useful to you.