LHC@home

I have just made a strange observation:

before, when I was crunching on Windows XP, a 2-core ATLAS task used about 16% of a 12-core CPU; this is a logical figure, right?

On the same machine (with dual boot) I now crunch 2-core ATLAS tasks on Windows 10, and - surprise - it uses about 24% of the same 12-core CPU (on former XP, this would have equaled a 3-core ATLAS task).

Which means that less CPU is left for other tasks.

Any explanation for this strange behaviour?

Your 12 core host running windows 10 successfully finished 2 ATLAS WUs today.
Both of them were 2-core VMs and show a CPU-time/runtime ratio of almost 2.
So, what's wrong?

https://lhcathome.cern.ch/lhcathome/result.php?resultid=203370240
https://lhcathome.cern.ch/lhcathome/result.php?resultid=203369055

This is a postponed:
2018-07-29 23:55:21 (7432): ERROR: Vboxwrapper lost communication with VirtualBox, rescheduling task for a later time.
Have this seen for me, when there is overloading. One CPU = One horsepower ;:)

This is a postponed ...

Yes, both of them. But at the end they finished successfully.
The OP was about the high CPU load which is not visible in the logs or timings.

I'm not shure, if my thaughts are correct, but:

If you limit a VM to a fixed number of cores, the VM will use only this number of cores to crunsh.

Let us now assume we are using a 2-Core-VM:

Inside the VM, only two cores ( = two tasks !) are used to crunch. But there are other tasks, that have to "support" these running crunsh-tasks and they need CPU-Power too. At this moment, when the box has idle cores, these tasks will get additive CPU-Cycles and this may lead to the situation, that a 2-Core-VM can use more than 2-Cores.for some time.

Remember that this is only my personal guess

---

Supporting BOINC, a great concept !

I'm not shure, if my thaughts are correct, but:

If you limit a VM to a fixed number of cores, the VM will use only this number of cores to crunsh.

Let us now assume we are using a 2-Core-VM:

Inside the VM, only two cores ( = two tasks !) are used to crunch. But there are other tasks, that have to "support" these running crunsh-tasks and they need CPU-Power too. At this moment, when the box has idle cores, these tasks will get additive CPU-Cycles and this may lead to the situation, that a 2-Core-VM can use more than 2-Cores.for some time.

Remember that this is only my personal guess

This would be a very serious hypervisor bug.
It's the job of the hypervisor (VirtualBox) to offer exactly the configured resources to the VM.
The OS inside the VM can't use more than this resources.

Additional cycles are used by the VirtualBox drivers, vboxwrapper and other auxiliary apps (if there are any) outside the VM.
The runtime and CPU time reported by BOINC are a result of the communication between BOINC client, vboxwrapper and VirtualBox.

As mentioned in other threads this may be the main reason why some versions of VirtualBox cause problems on some systems.

..
The runtime and CPU time reported by BOINC are a result of the communication between BOINC client, vboxwrapper and VirtualBox.
..

AFAIK since my early testing of BOINC's (ROM Walton's) vboxwrapper versions, the measured CPU-times is the sum of the cpu-times of the VBoxHeadless.exe's belonging to that VM-task.