climateprediction.net home page
AFlame PROJECT

AFlame PROJECT

Message boards : climateprediction.net Science : AFlame PROJECT
Message board moderation

To post messages, you must log in.

Previous · 1 · 2

AuthorMessage
Profile Alan K

Send message
Joined: 22 Feb 06
Posts: 484
Credit: 29,590,874
RAC: 1,482
Message 62621 - Posted: 21 Jul 2020, 22:31:52 UTC - in response to Message 62620.  

I recently had a task that was on its 4th go. All 3 previous attempts had failed within a few seconds due to missing libraries! Needless to say it completed on my machine.
ID: 62621 · Report as offensive     Reply Quote
Profile Dave Jackson
Volunteer moderator

Send message
Joined: 15 May 09
Posts: 4341
Credit: 16,496,276
RAC: 6,460
Message 62622 - Posted: 22 Jul 2020, 10:36:50 UTC - in response to Message 62621.  

I recently had a task that was on its 4th go. All 3 previous attempts had failed within a few seconds due to missing libraries! Needless to say it completed on my machine.


I have three tasks on my laptop at the moment. One is on its second attempt, the other two are on their fourth attempt. Out of the seven fails between the three, five are missing libraries. I forget what the others are.
ID: 62622 · Report as offensive     Reply Quote
Tut

Send message
Joined: 14 Dec 05
Posts: 27
Credit: 10,684,386
RAC: 1,967
Message 62623 - Posted: 23 Jul 2020, 22:04:15 UTC

Would it be possible to create a Linux test app that would merely verify (or not) the correct libraries on each downloading machine? It would then "somehow" notify that user that his client was (or not) configured correctly to process Linux tasks, and, if not, provide him with a link to explicit directions to remedy the situation. Part of the challenge would be to send it to every Linux box once and only once. This might also apply to Virtual Box installs. I frankly do not know enough about that except to wonder.
ID: 62623 · Report as offensive     Reply Quote
Profile Alan K

Send message
Joined: 22 Feb 06
Posts: 484
Credit: 29,590,874
RAC: 1,482
Message 62624 - Posted: 23 Jul 2020, 22:24:51 UTC - in response to Message 62620.  

Does the answer lie in the participant database? Computers without the libraries will fail the tsaks with a short CPU run time - maybe only a few seconds. Would it be possible to automatically block a computer that fails more than say 4 tasks within this time and message the owner to check his machine?
ID: 62624 · Report as offensive     Reply Quote
Profile geophi
Volunteer moderator

Send message
Joined: 7 Aug 04
Posts: 2167
Credit: 64,468,868
RAC: 3,594
Message 62625 - Posted: 24 Jul 2020, 3:21:15 UTC - in response to Message 62624.  
Last modified: 24 Jul 2020, 3:21:37 UTC

Does the answer lie in the participant database? Computers without the libraries will fail the tsaks with a short CPU run time - maybe only a few seconds. Would it be possible to automatically block a computer that fails more than say 4 tasks within this time and message the owner to check his machine?

Perhaps even easier, stderr.txt has a pretty set error message for the lack of 32bit libraries. New tasks being returned with that error message in stderr would flag that computer. Some or all of this perhaps

error while loading shared libraries: libstdc++.so.6: cannot open shared object file


It wouldn't get all 32bit library errors, but the vast majority of them.
ID: 62625 · Report as offensive     Reply Quote
Profile Alan K

Send message
Joined: 22 Feb 06
Posts: 484
Credit: 29,590,874
RAC: 1,482
Message 62626 - Posted: 24 Jul 2020, 22:34:28 UTC - in response to Message 62625.  
Last modified: 24 Jul 2020, 22:35:39 UTC

Yes, but does everyone who gets "computer error" next to a task check to see why there is an error? If they did then the same computers wouldn't be throwing up so many failures. The set it and forget it brigade.
ID: 62626 · Report as offensive     Reply Quote
Profile geophi
Volunteer moderator

Send message
Joined: 7 Aug 04
Posts: 2167
Credit: 64,468,868
RAC: 3,594
Message 62627 - Posted: 25 Jul 2020, 3:52:15 UTC - in response to Message 62626.  

I just meant that if one were to automate the task of finding computers crashing tasks due to 32 bit library deficiency, the message in stderr would be a good way to do it.
ID: 62627 · Report as offensive     Reply Quote
mikey

Send message
Joined: 18 Nov 18
Posts: 21
Credit: 5,536,552
RAC: 5,616
Message 62685 - Posted: 8 Sep 2020, 0:38:40 UTC - in response to Message 62626.  

Yes, but does everyone who gets "computer error" next to a task check to see why there is an error? If they did then the same computers wouldn't be throwing up so many failures. The set it and forget it brigade.


Less than 10% of Boinc users ever visit a projects forums to figure out what's going on, they just see it isn't working and either keep blindly trying or quit and go elsewhere.
ID: 62685 · Report as offensive     Reply Quote
mikey

Send message
Joined: 18 Nov 18
Posts: 21
Credit: 5,536,552
RAC: 5,616
Message 62686 - Posted: 8 Sep 2020, 0:39:45 UTC - in response to Message 62627.  

I just meant that if one were to automate the task of finding computers crashing tasks due to 32 bit library deficiency, the message in stderr would be a good way to do it.


I would think if you could write it the Project would at least look at it to see if it worked or not. Fixing the problem is another matter though.
ID: 62686 · Report as offensive     Reply Quote
mikey

Send message
Joined: 18 Nov 18
Posts: 21
Credit: 5,536,552
RAC: 5,616
Message 62687 - Posted: 8 Sep 2020, 1:03:00 UTC - in response to Message 62616.  

It's a pity that CPDN cant check for the libraries before downloading the task files or when LINUX boxes add CPDN as a project.


I agree. I asked about this over on the BOINC forums and no one there came up with ideas about how to do it. I also asked about including the libraries in the downloaded applications but it seems there would be problems with different Linux flavours and version numbers causing compatibility problems.


Why can't CPDN detect that it's not there and just add it to the CPDN Project folder and then set the app to look there if it can't find it in the usual place? ALL Boinc projects send a ton of files to us users when we sign up for the different projects just add the mising Libs.

CPDN could even find a set of Lib files that work with their app and tell the app to STOP using the ones found on the individuals computer and instead use the ones in the CPDN Project folders that are download when the person attaches to the Project. If the Lib files get changed on the computer who cares, CPDN has the ones it needs and if they need updating they can be sent by the Project.
ID: 62687 · Report as offensive     Reply Quote
lazlo_vii

Send message
Joined: 11 Dec 19
Posts: 108
Credit: 3,012,142
RAC: 0
Message 62690 - Posted: 8 Sep 2020, 2:58:09 UTC - in response to Message 62687.  
Last modified: 8 Sep 2020, 2:58:30 UTC

Why not use the "Notices" tab in the BOINC client to let them know when the libraries are missing? Another route could be to compile any needed 32bit code as a bunch of static libs and bundle them with the executables needed to run the WU's.
ID: 62690 · Report as offensive     Reply Quote
Profile Dave Jackson
Volunteer moderator

Send message
Joined: 15 May 09
Posts: 4341
Credit: 16,496,276
RAC: 6,460
Message 62692 - Posted: 8 Sep 2020, 7:51:18 UTC
Last modified: 8 Sep 2020, 8:00:42 UTC

Why can't CPDN detect that it's not there and just add it to the CPDN Project folder and then set the app to look there if it can't find it in the usual place? ALL Boinc projects send a ton of files to us users when we sign up for the different projects just add the missing Libs.

That route I am told would lead to incompatibility problems due to changes between releases, never mind different distributions. The project I am sure does not want to have the files for every single distribution out there!

Edit: Interestingly, on my new machine, I have got ten tasks from testing that did not crash despite my not having installed the missing libraries or at least not intentionally. I have just run the command to install them and got the message 32 packages to install and 1 not to upgrade. Presumably that one was the crucial one for the hadam4 tasks.
ID: 62692 · Report as offensive     Reply Quote
Previous · 1 · 2

Message boards : climateprediction.net Science : AFlame PROJECT

©2024 climateprediction.net