Message boards : Number crunching : Unusually high credit for Atlas tasks on one computer.
Message board moderation

To post messages, you must log in.

AuthorMessage
JHMarshall

Send message
Joined: 17 Jan 17
Posts: 2
Credit: 22,716,463
RAC: 0
Message 31548 - Posted: 22 Jul 2017, 4:56:40 UTC

I have one computer that is returning more than 10x the credits per Atlas tasks. Does anyone know what is happening here?

I have several other systems that are returning normal credit.



https://lhcathome.cern.ch/lhcathome/result.php?resultid=152030628
ID: 31548 · Report as offensive     Reply Quote
ivan
Volunteer moderator
Project tester
Volunteer developer
Volunteer tester
Project scientist
Avatar

Send message
Joined: 29 Aug 05
Posts: 998
Credit: 6,264,307
RAC: 71
Message 31551 - Posted: 22 Jul 2017, 10:15:09 UTC - in response to Message 31548.  

If you look at the Application Details for that machine, its GFLOP rating for Atlas is ~10x what it is for other apps. I don't know how that is determined, though.
ID: 31551 · Report as offensive     Reply Quote
gyllic

Send message
Joined: 9 Dec 14
Posts: 202
Credit: 2,533,875
RAC: 0
Message 31553 - Posted: 22 Jul 2017, 11:55:57 UTC - in response to Message 31548.  

i looked at the atlas logs from this host and, except for one tasks, the HITS.xxx files are missing.
this means that that most of the tasks from this host do NOT produce any physically (simulation) useable results and is so to say "wasting" the cpu resource.
normally the percentage of tasks with HITS file is much higher than the ones without. so maybe you were just unlucky with "getting bad tasks" or your host has some kind of a problem.
ID: 31553 · Report as offensive     Reply Quote
Toby Broom
Volunteer moderator

Send message
Joined: 27 Sep 08
Posts: 798
Credit: 644,758,739
RAC: 232,566
Message 31555 - Posted: 22 Jul 2017, 14:29:57 UTC

I think as the computer is new, the GFLOP isn't caculated correctly, once it returns somre results it will come down to normal level.
ID: 31555 · Report as offensive     Reply Quote
Eric Mcintosh
Volunteer moderator
Project administrator
Project developer
Project tester
Project scientist

Send message
Joined: 12 Jul 11
Posts: 857
Credit: 1,619,050
RAC: 0
Message 31557 - Posted: 22 Jul 2017, 17:10:53 UTC - in response to Message 31555.  

This is surely "my" SixTrack problem with NULL results for
whatever reason. This can now lead to tasks failing with
Real Time Limit Exceeded! Eric.


I think as the computer is new, the GFLOP isn't caculated correctly, once it returns somre results it will come down to normal level.

ID: 31557 · Report as offensive     Reply Quote
JHMarshall

Send message
Joined: 17 Jan 17
Posts: 2
Credit: 22,716,463
RAC: 0
Message 31559 - Posted: 22 Jul 2017, 18:13:51 UTC

One thing different on this computer is that it is Windows Server 2008 R2, while all my other systems are Windows 10. I don't want to crunch Atlas if it's not returning valid results. All the other apps appear to run normally. I'll try sixtrack for a while.

Note: the projected runtime for Atlas on this computer is way off. It started at a little over 4 minutes and has now increased to almost 8 minutes for newer WUs!!! The percent complete counts up normally to 100% at the 5-8 minute mark and then sits at 100% for 2 1/2 hours before the task completes.

There is also strange behavior with the CPU clock. If I run Atlas by itself the CPU is stuck at 1.6 GHz. It should be 3.5 GHz. If I run POGS by itself the clock kicks to 3.5 GHz as it should. If I run Atlas (clock 1.6GHz) and then start a POGS task the clock kicks up to 3.5 GHz. This is how I'm currently running to clear the Atlas cache (4 jobs). After that I'll switch to another LHC project to see how they work.
ID: 31559 · Report as offensive     Reply Quote
Profile Yeti
Volunteer moderator
Avatar

Send message
Joined: 2 Sep 04
Posts: 453
Credit: 193,369,412
RAC: 10,065
Message 31637 - Posted: 26 Jul 2017, 16:27:48 UTC - in response to Message 31559.  

... I don't want to crunch Atlas if it's not returning valid results.

Come with me on a sunny walk through this checklist


Supporting BOINC, a great concept !
ID: 31637 · Report as offensive     Reply Quote

Message boards : Number crunching : Unusually high credit for Atlas tasks on one computer.


©2024 CERN