LHC@home

Hi, I'm running 10 ATLAS simulations using 5950X on one computer.

The current state uses three threads. Allocate unlimited memory and disks.

If you increase the number of cores allocated per thread to more than 3 cores, is there a difference in processing time?

and

ATLAS is using too much memory.

Does it benefit from memory usage or work efficiency if you install Linux with vmware on your computer and run ATLAS in it?

There are lots of posts that explain how an ATLAS multicore setup works.
Some examples:
https://lhcathome.cern.ch/lhcathome/forum_thread.php?id=5486&postid=43258
https://lhcathome.cern.ch/lhcathome/forum_thread.php?id=5421&postid=42456
https://lhcathome.cern.ch/lhcathome/forum_thread.php?id=5169&postid=40670

It doesn't matter whether it happens inside a CERN VM or a self created VM.
Each core/thread you configure will need about 900 MB RAM.
In addition you need about 3 GB once per computer instance (bare metal/VM) to satisfy the needs of all other processes.
This includes the OS as well as the CVMFS cache.

Only one of your computers has enough RAM to run 10 ATLAS tasks concurrently (3-core setup):
https://lhcathome.cern.ch/lhcathome/show_host_detail.php?hostid=10692210

To allow this you would need to enable AMD-V.
This is from one of your logfiles:

VBoxManage.exe: error: Not in a hypervisor partition (HVP=0) (VERR_NEM_NOT_AVAILABLE).
VBoxManage.exe: error: AMD-V is disabled in the BIOS (or by the host OS) (VERR_SVM_DISABLED)
VBoxManage.exe: error: Details: code E_FAIL (0x80004005), component ConsoleWrap, interface IConsole

2021-06-26 16:52:40 (1552): VM failed to start.
2021-06-26 16:52:40 (1552): Could not start 
2021-06-26 16:52:40 (1552): ERROR: VM failed to start

I have read your link carefully.

1. There is no difference between running ATLAS directly on Linux OS or simulating ATLAS on Windows.

2. Formula for memory required for Atlas project (except Hyper-Threading)

5950x = 3000 + 14400MB (16core) = 17400MB

+@ 3GB of extra memory per vbox


3. Using as few cores as possible reduces overhead and increases efficiency (but ends faster with more cores)

Did I understand correctly?


p.s: AMD virtualization problem has already been solved :)

1. There is no difference between running ATLAS directly on Linux OS or simulating ATLAS on Windows.

The scientific apps and scripts are developed for Linux.
Even on a Linux host they usually run inside a singularity container to ensure the same kernel/library environment is used that the apps are developed for.
On a Windows host VirtualBox must be used to establish this Linux environment.

2. Formula for memory required for Atlas project (except Hyper-Threading)

The VirtualBox versions are limited to 8 cores => 10200 MB
The native version (NOT the extra large !) is limited to 12 cores => 13800 MB
The latter is less strict since the RAM is controlled by the host's OS.

Hyper-Threading/Threads/Cores
Often discussed but at the end very simple.
If HT is disabled in the BIOS your computer reports n CPUs to BOINC and VirtualBox.
If HT is enabled in the BIOS your computer reports 2*n CPUs to BOINC and VirtualBox.
Neither BOINC nor VirtualBox care about the internal hardware design of your CPU.

+@ 3GB of extra memory per vbox

That's already part of the formula.
The project server calculates:
3000 + 900 * n (with n = number of cores, limited to 8 or 12)

3. Using as few cores as possible reduces overhead and increases efficiency (but ends faster with more cores)

In most cases yes.
Using more cores per task usually allows to utilize more CPUs for ATLAS (in total) on a computer with less RAM.

Thank you very much for your kind reply.

It's been a lot of help.

Have a nice weekend. ^-^