Message boards : ATLAS application : ATLAS spends lots of time idling
Message board moderation

To post messages, you must log in.

AuthorMessage
Josu

Send message
Joined: 15 Mar 23
Posts: 2
Credit: 716,636
RAC: 0
Message 47872 - Posted: 19 Mar 2023, 22:36:42 UTC

Hello, everyone, I am new to LHC@home and have recently managed to get my Linux PC to successfully run work-units. I have installed VirtualBox for the VBox tasks, and CVMFS for the native tasks, and both are working okay. I would prefer running native tasks, as I have noticed slightly faster FLOPS as compared with VBox.

However, I have noticed that ATLAS Simulation jobs spend up to half an hour doing nothing before starting to use any CPU. The progress bar in BOINC Manager moves, but there is very little CPU or network usage, until later beginning to use the CPU, crunching for a while, waiting a little longer, and completing without errors. This happens with both VirtualBox and native tasks.

Is this normal behavior for ATLAS? What is it doing in that time? Only one ATLAS job will run at a time, and BOINC will not start anything else until the job is finished, as the ATLAS jobs use all CPU each. In other words, ATLAS is spending about 30 valuable minutes doing nothing for a while before it randomly decides to start working, then completing successfully. I have read that the firewall may cause issues, but since both VBox and Native tasks do eventually complete successfully after they start, I am not sure that is an issue.

System info:
AMD Ryzen 7 5800X (8-core CPU) and 32GB RAM
Ubuntu Studio 22.04 - Linux 5.15.0-67-lowlatency
BOINC 7.18.4 (Installed via APT)
VirtualBox 6.1 (Installed via APT)
ID: 47872 · Report as offensive     Reply Quote
maeax

Send message
Joined: 2 May 07
Posts: 2071
Credit: 156,141,325
RAC: 105,216
Message 47873 - Posted: 19 Mar 2023, 23:49:57 UTC - in response to Message 47872.  

Welcome Josu.
Atlas is running difficult. You have a successful Task finished,
Yeti's checklist in this Folder have a good background to find in the first step some news, how Atlas Task is working.
The beginning and ending of a Atlas Task need some time.
ID: 47873 · Report as offensive     Reply Quote
computezrmle
Volunteer moderator
Volunteer developer
Volunteer tester
Help desk expert
Avatar

Send message
Joined: 15 Jun 08
Posts: 2386
Credit: 222,953,824
RAC: 136,953
Message 47875 - Posted: 20 Mar 2023, 8:33:57 UTC - in response to Message 47872.  

Some notes


1. lowlatency kernel

Avoid it if you don't run applications where it is a must.
https://help.ubuntu.com/community/UbuntuStudio/RealTimeKernel

Don't run BOINC on a computer where a lowlatency kernel is a must.
And, yes I know that there are people claiming the lowlatency kernel being the "best" for normal users.


2. LHC@home apps

SixTrack:
Classical BOINC app, no special settings required

Vbox CMS/Theory:
1-core VMs, require VirtualBox (your v6.1 is OK)

native Theory:
requires a local CVMFS client, will fail on systems using cgroups v2 as default
https://lhcathome.cern.ch/lhcathome/result.php?resultid=389487566
https://lhcathome.cern.ch/lhcathome/forum_thread.php?id=5301&postid=47238

Vbox ATLAS:
8-core VMs (default on your computer, #cores/VM can be reduced, e.g. via app_config.xml)

native ATLAS:
requires a local CVMFS client
All configured cores are only used during event processing phase.
Startup and phase-out always run on 1 core
Startup (also on ATLAS vbox): prepares the runtime environment and updates the local CVMFS cache
This takes a while with little CPU activity. 20 min are not unusual.
In connection with (4., see below) startup takes much longer.


3. Website prefs

A Linux client will randomly get vbox and native tasks if "Run native if available?" is set.
To get native only "<dont_use_vbox>1</dont_use_vbox>" needs to be set in cc_config.xml.
Of course, then you will not get any CMS vbox tasks or vbox tasks from other projects.


4. CVMFS client

Although it delivers files, your client is not correctly configured.
[2023-03-19 14:08:20] 2.10.1.0 43467 181 60584 116937 2 63 12342959 16384000 0 130560 0 235405 99.129 1499589 4202 http://s1fnal-cvmfs.openhtc.io:8080/cvmfs/atlas.cern.ch http://131.225.188.245:3126 1
[2023-03-19 14:08:20] CVMFS is ok

This part "http://131.225.188.245:3126" shows that a backup proxy at Fermilab is used.
You may need to go through the steps mentioned here:
https://lhcathome.cern.ch/lhcathome/forum_thread.php?id=5594
https://lhcathome.cern.ch/lhcathome/forum_thread.php?id=5595
ID: 47875 · Report as offensive     Reply Quote
Crystal Pellet
Volunteer moderator
Volunteer tester

Send message
Joined: 14 Jan 10
Posts: 1268
Credit: 8,421,616
RAC: 2,139
Message 47876 - Posted: 20 Mar 2023, 8:36:23 UTC - in response to Message 47872.  

Is this normal behavior for ATLAS? What is it doing in that time? Only one ATLAS job will run at a time, and BOINC will not start anything else until the job is finished, as the ATLAS jobs use all CPU each.
Yes, this is normal ATLAS behaviour. It's loading data, libraries etc.
For the native tasks, it could be useful to run subsequent ATLAS-tasks using cached data. I'm not running native tasks myself.
For VBox tasks you could try to setup 4-core tasks in your project preferences in stead of 8-core.
Memory and cpu-cycles could be better used, specially when you don't start the 2 ATLAS-tasks at the same time, but with an interval of half an hour.
Towards the end of a task the result HITS-file is created, not using much cpu, but takes some time.

The progress BOINC is showing is rather useless.
ID: 47876 · Report as offensive     Reply Quote
Josu

Send message
Joined: 15 Mar 23
Posts: 2
Credit: 716,636
RAC: 0
Message 47877 - Posted: 20 Mar 2023, 12:00:21 UTC - in response to Message 47875.  

Thanks for the help! As some of the audio programs I use say they require the lowlatency kernel, I'll see if I can use the regular kernal and have them still work properly. As for CVMFS, I will make sure it is conigured properly.
ID: 47877 · Report as offensive     Reply Quote

Message boards : ATLAS application : ATLAS spends lots of time idling


©2024 CERN