climateprediction.net home page
no climate prediciton on debian etch ?

no climate prediciton on debian etch ?

Questions and Answers : Unix/Linux : no climate prediciton on debian etch ?
Message board moderation

To post messages, you must log in.

AuthorMessage
old_user493497

Send message
Joined: 5 Jan 08
Posts: 1
Credit: 0
RAC: 0
Message 32017 - Posted: 5 Jan 2008, 16:28:39 UTC

I had the problem that I always had the error :
Output file bla.zip for task bla absent

There was no error message in the logs.
So I made an strace and found this :
read(8, \"hadsm3_5.10_x86_64-pc-linux-gnu: /lib32/libc.so.6: version `GLIBC_2.4\\\' not found (required by hadsm3_5.10_x86_64-pc-linux-gnu)\\n\", 4096) = 127
close(8)

debian etch however has glicbc 2.3.

is there a way to get it working on debian etch ?

On my laptop it runs nicely though , its just my server that runs debian etch ( the amd64 )

Greetings,
josi

( sorry about making 3 accounts now :/ )
ID: 32017 · Report as offensive     Reply Quote
Profile geophi
Volunteer moderator

Send message
Joined: 7 Aug 04
Posts: 2167
Credit: 64,403,322
RAC: 5,085
Message 32019 - Posted: 5 Jan 2008, 19:02:17 UTC - in response to Message 32017.  

So I made an strace and found this :
read(8, \"hadsm3_5.10_x86_64-pc-linux-gnu: /lib32/libc.so.6: version `GLIBC_2.4\\\' not found (required by hadsm3_5.10_x86_64-pc-linux-gnu)\\n\", 4096) = 127
close(8)

debian etch however has glicbc 2.3.

is there a way to get it working on debian etch ?

It would appear not.

http://www.linuxquestions.org/questions/debian-26/debian-lenny-glibc2.4-missing-556712/
http://forums.debian.net/viewtopic.php?p=64434#64434
ID: 32019 · Report as offensive     Reply Quote
Les Bayliss
Volunteer moderator

Send message
Joined: 5 Sep 04
Posts: 7629
Credit: 24,240,330
RAC: 0
Message 32020 - Posted: 5 Jan 2008, 19:03:43 UTC
Last modified: 5 Jan 2008, 19:09:00 UTC

You\'ll need to get hold of GLIBC_2.4 and put it in the same place as GLIBC_2.3

Note:
If you have multiple accounts, the credits and other data CANNOT be merged together.

edit
But apparently Etch doesn\'t have, and never will have, 2.4
So you\'ll need to change the OS to one that DOES have it, or don\'t run climate models on the server.


Backups: Here
ID: 32020 · Report as offensive     Reply Quote
old_user493574

Send message
Joined: 5 Jan 08
Posts: 5
Credit: 0
RAC: 0
Message 32024 - Posted: 5 Jan 2008, 22:25:49 UTC - in response to Message 32020.  

Yes I came to that conclusion too. I calculate quantums now.

I wont update Glibc just to let my server do stuff while beeing idle .

I though there might be a version that could work with an earlier version of glibc.

Greetings ,
Josi
ID: 32024 · Report as offensive     Reply Quote
Profile MikeMarsUK
Volunteer moderator
Avatar

Send message
Joined: 13 Jan 06
Posts: 1498
Credit: 15,613,038
RAC: 0
Message 32027 - Posted: 6 Jan 2008, 1:17:00 UTC
Last modified: 6 Jan 2008, 1:20:04 UTC

There are three different models, so you could try each in turn to see what versions of glibc are required. The instructions to do this are in the README \'running your model\'.

Having said that, I wouldn\'t be entirely surprised if the requirements were the same on each one (I\'d guess they were built on the same machine, although I don\'t know for sure).

I was under the impression that the more recent versions (5.8+) of Boinc itself also requires glibc 2.4, or is this no longer the case?
I'm a volunteer and my views are my own.
News and Announcements and FAQ
ID: 32027 · Report as offensive     Reply Quote
Profile geophi
Volunteer moderator

Send message
Joined: 7 Aug 04
Posts: 2167
Credit: 64,403,322
RAC: 5,085
Message 32028 - Posted: 6 Jan 2008, 5:37:28 UTC

I was running a hadsm3 5.06 under 32bit BOINC 5.10.21 on Fedora Core 4 which has glibc_2.3. But that\'s 32bit and you\'re running the 64bit BOINC client. You might have some luck going to the 32bit BOINC client.
ID: 32028 · Report as offensive     Reply Quote
Profile MikeMarsUK
Volunteer moderator
Avatar

Send message
Joined: 13 Jan 06
Posts: 1498
Credit: 15,613,038
RAC: 0
Message 32033 - Posted: 6 Jan 2008, 11:23:12 UTC
Last modified: 6 Jan 2008, 11:23:34 UTC


It\'s amazing that such a recent distribution (kernal 2.6.22 and 64-bit O/S) doesn\'t have the up-to-date system libraries included within the distribution! I guess a server doesn\'t need graphics...

I'm a volunteer and my views are my own.
News and Announcements and FAQ
ID: 32033 · Report as offensive     Reply Quote
old_user493574

Send message
Joined: 5 Jan 08
Posts: 5
Credit: 0
RAC: 0
Message 32133 - Posted: 11 Jan 2008, 17:01:52 UTC - in response to Message 32033.  


It\'s amazing that such a recent distribution (kernal 2.6.22 and 64-bit O/S) doesn\'t have the up-to-date system libraries included within the distribution! I guess a server doesn\'t need graphics...



No the PC you see now on my account is my laptop, which runs debian lenny and has glibc 2.7, I \"deleted\" ( changed the email) of my 1st account . my server runs kernel 2.6.18 . I backported Boinc 5.10 from the debian unstable repositories on my server without any problems . ( I 1st thought boinc 5.4, from stable aka etch, would be too old for this project )

I guess the other tasks I can get here are based glib c 2.4 + so I didnt try ( I can only make 2 requests a day so i cant even try them all ^^ )

ID: 32133 · Report as offensive     Reply Quote
DJStarfox

Send message
Joined: 27 Jan 07
Posts: 300
Credit: 3,288,263
RAC: 26,370
Message 32140 - Posted: 12 Jan 2008, 3:33:58 UTC - in response to Message 32133.  

No the PC you see now on my account is my laptop, which runs debian lenny and has glibc 2.7, I \"deleted\" ( changed the email) of my 1st account . my server runs kernel 2.6.18 . I backported Boinc 5.10 from the debian unstable repositories on my server without any problems . ( I 1st thought boinc 5.4, from stable aka etch, would be too old for this project )

I guess the other tasks I can get here are based glib c 2.4 + so I didnt try ( I can only make 2 requests a day so i cant even try them all ^^ )


Oh. The GLIBC 2.4 requirement started with the 5.10.x releases. Did you try 5.8.16?
ID: 32140 · Report as offensive     Reply Quote
old_user493574

Send message
Joined: 5 Jan 08
Posts: 5
Credit: 0
RAC: 0
Message 32146 - Posted: 12 Jan 2008, 11:33:58 UTC - in response to Message 32140.  


Oh. The GLIBC 2.4 requirement started with the 5.10.x releases. Did you try 5.8.16?


nope:
boinc --version
5.10.30 x86_64-pc-linux-gnu

And I\'m sure I have glibc 2.3 ^^.
I compiled the client myself though.

The client works fine , it\'s just that climate prediction needs glibc 2.4 .
Or would I get a Glibc 2.3 compiled version of climate prediciton if i used boinc 5.8 ?
ID: 32146 · Report as offensive     Reply Quote
Profile MikeMarsUK
Volunteer moderator
Avatar

Send message
Joined: 13 Jan 06
Posts: 1498
Credit: 15,613,038
RAC: 0
Message 32147 - Posted: 12 Jan 2008, 12:20:57 UTC


It\'s separate - both the boinc manager and the climate code will want to load libraries, and the versions may not be the same.

I'm a volunteer and my views are my own.
News and Announcements and FAQ
ID: 32147 · Report as offensive     Reply Quote
old_user493574

Send message
Joined: 5 Jan 08
Posts: 5
Credit: 0
RAC: 0
Message 32154 - Posted: 12 Jan 2008, 20:20:37 UTC - in response to Message 32147.  


It\'s separate - both the boinc manager and the climate code will want to load libraries, and the versions may not be the same.


Well I manage my Server remotely . so the manager runs on my laptop .

Ok Conclusion is : To run this project I need Glibc 2.4 , which I don\'t have .
See you once Debian Lenny goes stable :).

ID: 32154 · Report as offensive     Reply Quote
Palo M.
Avatar

Send message
Joined: 8 Dec 04
Posts: 2
Credit: 1,459,257
RAC: 0
Message 33204 - Posted: 4 Apr 2008, 19:04:41 UTC - in response to Message 32154.  

Ok Conclusion is : To run this project I need Glibc 2.4 , which I don\'t have .
See you once Debian Lenny goes stable :).


I\'m in the same situation: Having 64-bit Debian stable (Etch) on Q6600 (with 4 GB RAM, so it does not make sense to run 32-bit system there). Machine runs 24/7 (it is server, that\'s why it is running stable distribution and not unstable), so it is very suitable for CPDN which takes long time... I was optimistic with this new machine, but all I\'ve got is failed WUs.
And as I read on several places of the forum, it is not going to be solved soon - so it seems project does not care about such machines (which sounds strange to me, since memory consumption by CPDN applications is big and 32-bit machine has memory limit).
If CPDN doesn\'t want my Q6600, let it be so. There are several projects which send 64-bit applications directly and some other projects are open-source, so we can compile applications ourselves.

BTW, I do not understand, why dynamically linked applications are distributed and not static - dynamic linking has just big chance it will not work on some systems. It is the same with BOINC-core-client - official Linux 64-bit binary is dynamically linked (with no static alternative) and it needs GLIBC_2.4, so it didn\'t work for me; I had to compile it myself (which went smoothly without any problems on the other hand).
There is CPDN-beta - according to this post it uses glibc 2.3 and my first WU didn\'t crash yet - but it\'s still dynamically linked 32-bit application, so it is not ideal for 64-bit systems as well. I will let this WU run, but I will probably not start another until there\'s 64-bit...
ID: 33204 · Report as offensive     Reply Quote
Profile astroWX
Volunteer moderator

Send message
Joined: 5 Aug 04
Posts: 1496
Credit: 95,522,203
RAC: 0
Message 33205 - Posted: 4 Apr 2008, 19:24:15 UTC

The 32-bit vs. 64-bit issue comes up time and again. An attempt to port to 64-bit was attempted a couple years ago. It died in Alpha because there was not staff resource available to support it. If memory serves, there was no performance gain to justify the effort.

Static-linked Models were distributed at one point. (Perhaps it was only in Alpha or Beta test. I don\'t recall why the change was made.)

Forget about compiling the million lines of Fortran which constitute CPDN, it doesn\'t belong to Oxford; the code is proprietary and belongs to the UK Met. Office and is used under license. Besides, any tweaks would require a horrendous amount of scientific validation before the results could be used.

"We have met the enemy and he is us." -- Pogo
Greetings from coastal Washington state, the scenic US Pacific Northwest.
ID: 33205 · Report as offensive     Reply Quote
Les Bayliss
Volunteer moderator

Send message
Joined: 5 Sep 04
Posts: 7629
Credit: 24,240,330
RAC: 0
Message 33206 - Posted: 4 Apr 2008, 20:27:10 UTC


... so it seems project does not care about such machines ...


People keep saying this, but the truth lies elsewhere: people keep running operating systems or hardware that don\'t have the software needed by this project.

The program used here is mostly floating point, whereas a lot of the others are mostly integer, making them better suited to amateurs rearranging the code to \'improve\' it. Which may be OK when all it does is check a list of data against certain expected values.

Here though, the program is starting from a set of values, and then creating a 3D model of the Earth\'s atmosphere and ocean. So anyone changing the code needs to understand climatology. And then, as astroWX said, the results need to be validated against known results for those same starting values.
A lot of attempts have been made to improve the speed, and to make the program work better than it currently is on certain cpus, but each time the models have turned out to be useless.


Backups: Here
ID: 33206 · Report as offensive     Reply Quote
Palo M.
Avatar

Send message
Joined: 8 Dec 04
Posts: 2
Credit: 1,459,257
RAC: 0
Message 33208 - Posted: 5 Apr 2008, 5:33:56 UTC - in response to Message 33206.  

Thank you for your explanations. I understand that compilation of CPDN applications is more complex than usual compilation of other application...
BTW, it is possible to link statically some libraries while having some others linked dynamically (i.e. standard libs can be linked statically to avoid incompatibility, and \"own\" libraries can be linked dynamically, as they are distributed by the project anyway).

...
The 32-bit vs. 64-bit issue comes up time and again. An attempt to port to 64-bit was attempted a couple years ago. It died in Alpha because there was not staff resource available to support it. If memory serves, there was no performance gain to justify the effort.

In my case it was not about performance - I agree that performance cannot be improved that much - and little shorter (let\'s say 10%) computation time isn\'t worth the effort; more important is that models are computed till the end.
In my case it is about ability (or better said inability) to run CPDN at all.

And, 32-bit Debian etch uses the same glibc 2.3.6, so probably the same problem would happen there anyway (so maybe it\'s not problem of 64-bit at all, just glibc incompatibility problem).

...
People keep saying this, but the truth lies elsewhere: people keep running operating systems or hardware that don\'t have the software needed by this project.

I run 64-bit system for the reason - 32-bit system cannot use all 4G memory that I have (well, custom 32-bit Linux kernel could do it in principle, but it would be too slow, so it wouldn\'t make sense). It is the server, therefore 4G and therefore stable distribution (and, I\'m not going to overclock it).
It seems Josi is in exactly same situation as me - maybe we are only two, or maybe there are more of us, I don\'t know. But Debian is one of the most popular distributions.
BTW, it\'s pretty sure that the world is moving towards 64 bits - as soon as someone uses 4G or more, 64-bit OS is the only reasonable way. And x86-64 architecture has big advantage of backward compatibility with i386 - so this transition needn\'t be very painful. I see that project did something in this direction: i386 version of HadSM is provided also for x86-64 architecture - it\'s just pity it doesn\'t run on some systems due to libraries incompatibility... such a small step is missing. I know it\'s not easy because of complicated verification; I hope CPDN beta could help here.
ID: 33208 · Report as offensive     Reply Quote
Les Bayliss
Volunteer moderator

Send message
Joined: 5 Sep 04
Posts: 7629
Credit: 24,240,330
RAC: 0
Message 33209 - Posted: 5 Apr 2008, 7:54:51 UTC
Last modified: 5 Apr 2008, 7:58:06 UTC

64 bit OS is indeed the way foward, but not for the models currently available here. They are being reused, with variations, for related projects. e.g. the data from the current round of slab models is for a German project.
It took about 2 years to port the current program from the supercomputer version into a desktop version, and to get it to run stably. It\'s unlikely that this will happen again to create a 64 bit specific version, due to the present large increase of work for the 2 project people.

The \'64 bit\' slab models are just 32 bit models, with the server software recognizing when a 64 bit computer asks for a model.

There was, and possibly still is, a project at another UK Uni for hi-resolution models.
These will need 4.5 Gb for the model alone, so around 6 gigs for model plus OS.

ID: 33209 · Report as offensive     Reply Quote
Profile astroWX
Volunteer moderator

Send message
Joined: 5 Aug 04
Posts: 1496
Credit: 95,522,203
RAC: 0
Message 33213 - Posted: 5 Apr 2008, 21:16:04 UTC

Five of my boxes are dual boot XP/openSuSE x64.* Fortunately, for me, Yast2 loads all the necessary libraries to run 32-bit apps without issues (well, not counting driver and convenience issues), that includes CPDN.

I think it\'s necessary to pound on Ubuntu/Kubuntu to get the necessary libraries installed, too. Luck of the draw with one one\'s choice of distro, I suppose. (As I see it, that\'s a major impediment to getting more people to use Linux.)

Debian was the CPDN development vehicle for Linux, I think. Not sure that\'s still true.

* For anyone considering Vista/Linux dual boot, a tweak of the boot loader is necessary after adding Linux -- thanks to a change in NTFS. (M$ strikes again!)
"We have met the enemy and he is us." -- Pogo
Greetings from coastal Washington state, the scenic US Pacific Northwest.
ID: 33213 · Report as offensive     Reply Quote
Desti

Send message
Joined: 6 Aug 04
Posts: 124
Credit: 9,195,838
RAC: 0
Message 33225 - Posted: 6 Apr 2008, 17:17:59 UTC - in response to Message 33209.  

[...]

There was, and possibly still is, a project at another UK Uni for hi-resolution models.
These will need 4.5 Gb for the model alone, so around 6 gigs for model plus OS.



I hope they are still working on this, I\'m sure 2010 every new standard PC will have 6-8 GiB RAM, 8 core processor and 1-2 TB diskspace.
Linux Users Everywhere @ BOINC
ID: 33225 · Report as offensive     Reply Quote

Questions and Answers : Unix/Linux : no climate prediciton on debian etch ?

©2024 climateprediction.net