Errors on Linux x64

log in

Advanced search

Message boards : Number crunching : Errors on Linux x64

Author Message
Chris Granger
Send message
Joined: 2 Dec 10
Posts: 12
Credit: 88,661
RAC: 0
Message 677 - Posted: 8 May 2011, 23:56:20 UTC

I'm afraid that tasks are erroring out right as they begin, on my 64-bit Ubuntu rig running Gibbs sampler v0.14... Any idea what might be wrong?

My Windows machine is working just fine with Gibbs sampler v0.18 though, so that's good.

kashi
Send message
Joined: 9 Apr 11
Posts: 29
Credit: 526,885
RAC: 0
Message 678 - Posted: 9 May 2011, 1:08:56 UTC

BOINC is sending many 32-bit v0.14 tasks that are incompatible with your 64-bit Linux. The correct version v0.13 tasks for your 64-bit Linux are completing successfully on your computer.

This is a design error of BOINC. It has been mentioned in other threads.

Chris Granger
Send message
Joined: 2 Dec 10
Posts: 12
Credit: 88,661
RAC: 0
Message 679 - Posted: 9 May 2011, 6:11:43 UTC - in response to Message 678.

I see. So there's nothing to be done but wait for a bug fix?

kashi
Send message
Joined: 9 Apr 11
Posts: 29
Credit: 526,885
RAC: 0
Message 682 - Posted: 9 May 2011, 10:58:24 UTC - in response to Message 679.
Last modified: 9 May 2011, 11:07:25 UTC

Perhaps you could "update your gcc" http://dnahome.cs.rpi.edu/dna/forum_thread.php?id=87&nowrap=true#659
That's Linux talk, I think it means update your operating system.

I remember when doing QMC in 64-bit Linux you had to add ia32-libs package plus some others to allow processing the QMC 32-bit Linux application. Those more familiar with Linux may be able to answer more knowledgably about this.
The Dotsch/UX Linux I use for BOINC so far has had no trouble processing 32-bit BOINC applications.

Another way is to use an app_info.xml file which should prevent BOINC from sending 32-bit tasks. I do not know how to do this for this project.

Getting numerous errors or aborting tasks should possibly be avoided on any project using the BOINC assigned credit scheme. I'm not sure but it may affect the amount of credit granted. If any component of the scheme is based on the BOINC benchmark then a higher benchmark relative to task runtime is also desirable for you and your wingman. This may be advantageous to those with hyper threading CPUs and also CPUs with turbo boost features.

Travis Desell
Volunteer moderator
Project administrator
Project developer
Project tester
Project scientist
Send message
Joined: 8 Feb 10
Posts: 364
Credit: 262,678
RAC: 0
Message 683 - Posted: 9 May 2011, 11:21:21 UTC - in response to Message 682.

Perhaps you could "update your gcc" http://dnahome.cs.rpi.edu/dna/forum_thread.php?id=87&nowrap=true#659
That's Linux talk, I think it means update your operating system.

I remember when doing QMC in 64-bit Linux you had to add ia32-libs package plus some others to allow processing the QMC 32-bit Linux application. Those more familiar with Linux may be able to answer more knowledgably about this.
The Dotsch/UX Linux I use for BOINC so far has had no trouble processing 32-bit BOINC applications.

Another way is to use an app_info.xml file which should prevent BOINC from sending 32-bit tasks. I do not know how to do this for this project.

Getting numerous errors or aborting tasks should possibly be avoided on any project using the BOINC assigned credit scheme. I'm not sure but it may affect the amount of credit granted. If any component of the scheme is based on the BOINC benchmark then a higher benchmark relative to task runtime is also desirable for you and your wingman. This may be advantageous to those with hyper threading CPUs and also CPUs with turbo boost features.


If he's getting sent the 32 bit binary to his 64 bit linux system, there's not much that can be done except me getting some kind of fix from the BOINC lists. It's pretty strange it's sending out a different version of the application.

Profile skgiven
Avatar
Send message
Joined: 3 May 11
Posts: 31
Credit: 827,977
RAC: 0
Message 686 - Posted: 9 May 2011, 11:28:39 UTC - in response to Message 683.

I'm not sure why you are using both 64bit and 32bit 0.17 app versions for both Win and Linux, or if it would help to separate them by version number.
For example,
Win x86 Version 0.17, x64 Version 0.18
32bit Linux 0.19, 64bit Linux 0.20

I can't see anything on the tasks properties to indicate whether they are either x86 or x64. Might be useful.

Profile Cori
Avatar
Send message
Joined: 29 Apr 10
Posts: 108
Credit: 804,178
RAC: 0
Message 703 - Posted: 9 May 2011, 16:55:38 UTC - in response to Message 682.
Last modified: 9 May 2011, 17:22:27 UTC

... Another way is to use an app_info.xml file which should prevent BOINC from sending 32-bit tasks. I do not know how to do this for this project...

I just tried as well but I am always getting "unknown app name in your app_info.xml". :-(


EDIT: Now BOINC doesn't complain anymore about my app_info.xml but I don't get WUs. *LOL*
____________
Grrrrrrrrreetings from the Lazy Cat...

Profile Cori
Avatar
Send message
Joined: 29 Apr 10
Posts: 108
Credit: 804,178
RAC: 0
Message 705 - Posted: 9 May 2011, 17:57:24 UTC

Finally I got the app_info.xml's working for Windows and Linux x64! :-)))
____________
Grrrrrrrrreetings from the Lazy Cat...

zombie67 [MM]
Avatar
Send message
Joined: 25 Apr 10
Posts: 34
Credit: 1,012,863
RAC: 0
Message 716 - Posted: 10 May 2011, 4:05:07 UTC

Well go on then! Post them for us lazy bones! ;)
____________
Dublin, CA
Team SETI.USA

Profile Cori
Avatar
Send message
Joined: 29 Apr 10
Posts: 108
Credit: 804,178
RAC: 0
Message 719 - Posted: 10 May 2011, 10:42:59 UTC - in response to Message 716.

And there I though I was a lazy cat... *big grin*
So here we go for Windows x64:

<app_info>

<app>
<name>Gibbs</name>
<user_friendly_name>Gibbs sampler</user_friendly_name>
</app>

<file_info>
<name>Gibbs_0.18_windows_x86_64.exe</name>
<executable/>
</file_info>

<file_info>
<name>mtub_NC_000962_intergenic.2.43X.fa</name>
<sticky/>
</file_info>

<app_version>
<app_name>Gibbs</app_name>
<version_num>18</version_num>
<file_ref>
<file_name>Gibbs_0.18_windows_x86_64.exe</file_name>
<main_program/>
</file_ref>
</app_version>

</app_info>




And for Linux x64:

<app_info>

<app>
<name>Gibbs</name>
<user_friendly_name>Gibbs sampler</user_friendly_name>
</app>

<file_info>
<name>Gibbs_0.18_x86_64-pc-linux-gnu</name>
<executable/>
</file_info>

<file_info>
<name>mtub_NC_000962_intergenic.2.43X.fa</name>
<sticky/>
</file_info>

<app_version>
<app_name>Gibbs</app_name>
<version_num>18</version_num>
<file_ref>
<file_name>Gibbs_0.18_x86_64-pc-linux-gnu</file_name>
<main_program/>
</file_ref>
</app_version>

</app_info>

____________
Grrrrrrrrreetings from the Lazy Cat...

Travis Desell
Volunteer moderator
Project administrator
Project developer
Project tester
Project scientist
Send message
Joined: 8 Feb 10
Posts: 364
Credit: 262,678
RAC: 0
Message 731 - Posted: 10 May 2011, 22:58:21 UTC - in response to Message 719.

Hopefully using that won't make all your rigs break when I update the applications again (there still are a lot of things that still need to be implemented).

Profile Cori
Avatar
Send message
Joined: 29 Apr 10
Posts: 108
Credit: 804,178
RAC: 0
Message 733 - Posted: 11 May 2011, 7:34:39 UTC - in response to Message 731.

Hehe, no... only 3 boxes in total - I should be able to manage that. *grin*
____________
Grrrrrrrrreetings from the Lazy Cat...

zombie67 [MM]
Avatar
Send message
Joined: 25 Apr 10
Posts: 34
Credit: 1,012,863
RAC: 0
Message 735 - Posted: 11 May 2011, 23:25:35 UTC

It looks like the project-side fix worked. So the need is gone, I think.
____________
Dublin, CA
Team SETI.USA


Post to thread

Message boards : Number crunching : Errors on Linux x64


Main page · Your account · Message boards


Copyright © 2014 Travis Desell, Lee Newberg, Malik Magdon-Ismail and Boleslaw K. Szymanski