Quantcast

Parallel execution

classic Classic list List threaded Threaded
10 messages Options
christian.wohlschlager@jku.at christian.wohlschlager@jku.at
Reply | Threaded
Open this post in threaded view
|  
Report Content as Inappropriate

Parallel execution

Hi !


to my excuse I'm new in this.

I want to use scilab on my SGI UV 1000  with 2000 Cores ,i download it (the binary version)  everything ok, the parallel option seems ??? to be installed so i try it but while i monitor the cpu i coudn't find out that scilab is using more than 1 CPU.

even the test says everything computet in different threads but ????.

Did you have any Answere what i'm doing wrong, we just wanted to change to scilab because matlab  wantet to charge us for 1000 cpu workers 70 000 Euro !!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!! and telling us that Matlab is soo extremly performant just on this Supercomputer . Thats wy we want to change.


thank you


Christian Wohlschlager     


Christian Wohlschlager
JKU Universität Linz
Altenbergerstr.69
4040 Linz
Austria



Samuel GOUGEON Samuel GOUGEON
Reply | Threaded
Open this post in threaded view
|  
Report Content as Inappropriate

Re: Parallel execution

Hello,

Which OS are you using? On Windows, parallel_run() runs only 1 processor.
It is a limitation documented in the parallel_run help page.

Best regards
Samuel Gougeon


Le 16/11/2011 17:34, Christian Wohlschlager a écrit :

Hi !


to my excuse I'm new in this.

I want to use scilab on my SGI UV 1000  with 2000 Cores ,i download it (the binary version)  everything ok, the parallel option seems ??? to be installed so i try it but while i monitor the cpu i coudn't find out that scilab is using more than 1 CPU.

even the test says everything computet in different threads but ????.

Did you have any Answere what i'm doing wrong, we just wanted to change to scilab because matlab  wantet to charge us for 1000 cpu workers 70 000 Euro !!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!! and telling us that Matlab is soo extremly performant just on this Supercomputer . Thats wy we want to change.


thank you


Christian Wohlschlager     


Christian Wohlschlager
JKU Universität Linz
Altenbergerstr.69
4040 Linz
Austria




christian.wohlschlager@jku.at christian.wohlschlager@jku.at
Reply | Threaded
Open this post in threaded view
|  
Report Content as Inappropriate

Re: Parallel execution

Sorry for giving you not much (any ?) Details My OS is Novelle Suse SLES 11
Compiler I Have GnuCC/4.1.2    GnuCC/4.4

icc 12.0.3
ifort 12.0.3

i like to link my intel mkl 2011 libraries with it so the performance will be much better.

thank you
christian

christian.wohlschlager@jku.at christian.wohlschlager@jku.at
Reply | Threaded
Open this post in threaded view
|  
Report Content as Inappropriate

Re: Parallel execution

This post has NOT been accepted by the mailing list yet.
In reply to this post by Samuel GOUGEON
Sorry for giving you not much (any ?) Details My OS is Novelle Suse SLES 11
Compiler I Have GnuCC/4.1.2    GnuCC/4.4

icc 12.0.3
ifort 12.0.3

i like to link my intel mkl 2011 libraries with it so the performance will be much better.

thank you
christian

christian.wohlschlager@jku.at christian.wohlschlager@jku.at
Reply | Threaded
Open this post in threaded view
|  
Report Content as Inappropriate

Re: Parallel execution

Hello Again !

Today i ran the example programm (the simple one with
function a=g(arg1)
  a=arg1*arg1
endfunction

and
tic;res=parallel_run(1:2000, g);toc

while doing this i start top -u<user>
i couldnt see any parallel runs
so i start again 3 times

the result was that the computer did't answere any more , we found out that more than 9000 !!! processes iwere running we had to start the machine again.(we have a SGI UV 1000)
I think the problem was that he started with nb_workers=0 which means take every cpu and this 3 times.
Thats my first clue am i right. how default to start with 8 provessors ?
i also think that the command top didnt show me the right thing.



christian



Samuel GOUGEON Samuel GOUGEON
Reply | Threaded
Open this post in threaded view
|  
Report Content as Inappropriate

Re: Parallel execution

Hello,

Le 18/11/2011 07:45, [hidden email] a écrit :
> .../...
> I think the problem was that he started with nb_workers=0 which means take
> every cpu and this 3 times.
> Thats my first clue am i right. how default to start with 8 provessors ?
>    
You may try with:

params = mlist(["plist" "nb_workers"],8);
tic;res=parallel_run(1:2000, g, params);toc

Samuel
sylvestre sylvestre
Reply | Threaded
Open this post in threaded view
|  
Report Content as Inappropriate

Re: Parallel execution

In reply to this post by christian.wohlschlager@jku.at
Hello Christian,

Le mercredi 16 novembre 2011 à 17:34 +0100, Christian Wohlschlager a
écrit :
> Hi !
>

>
> I want to use scilab on my SGI UV 1000  with 2000 Cores ,i download it
> (the binary version)  everything ok, the parallel option seems ??? to
> be installed so i try it but while i monitor the cpu i coudn't find
> out that scilab is using more than 1 CPU.
[..]

It depends what you want to do.
Scilab has several ways to use this kind of architecture:
* As Samuel said, parallel_run is one way [0]
* Under Linux, we do no ship with MKL, Gotoblas neither Atlas (but with
a very slow refblas). If you switch to one of them you will get both
great performances and all core being used during matrix computations.
* GPGPU: you can give a try to our GPGPU extensions (don't hesitate if
you have any feedback). [1]
* If you are more into grid/cloud computing, you can give a try to proactive
[2]
* In Scilab, even it is a bit outdated, you will find a PVM interface [3].
* Finally, if you are familiar on the build process of an linux application,
you can try with the MPI implementation of Scilab (in the git branch "MPI")

Hope this helps,
Sylvestre
[0] http://help.scilab.org/parallel_run
[1] http://atoms.scilab.org/toolboxes/sciGPGPU
[2] http://proactive.inria.fr/
[3] http://help.scilab.org/pvm

christian.wohlschlager@jku.at christian.wohlschlager@jku.at
Reply | Threaded
Open this post in threaded view
|  
Report Content as Inappropriate

Re: Parallel execution

Dear Sylvestre !


Thank you for the reply I'm using scilab on a SGI UV 1000 with 2000 cores

I try the example programm  with no params that's why i chrashed , yes ok, but how can i prevent the users to do the sane as i did . Currently i try with limit maxproec 128  (not 16 000 as default parameter)  on the command line.I didnt test it right now becouse i have to be sure that nobody is on the machine exept me otherwise :-( my boss will cut me into pieces.

Do you have aby other sussestion ??


thank you


Christian




Christian Wohlschlager
JKU Universität Linz
Altenbergerstr.69
4040 Linz
Austria



>>> "sylvestre [via Scilab / Xcos - Mailing Lists Archives]" <ml-node+[hidden email]> 11/22/2011 12:59 AM >>>
Hello Christian,

Le mercredi 16 novembre 2011 Ã 17:34 +0100, Christian Wohlschlager a
écrit :
> Hi !
>

>
> I want to use scilab on my SGI UV 1000  with 2000 Cores ,i download it
> (the binary version)  everything ok, the parallel option seems ??? to
> be installed so i try it but while i monitor the cpu i coudn't find
> out that scilab is using more than 1 CPU.
[..]

It depends what you want to do.
Scilab has several ways to use this kind of architecture:
* As Samuel said, parallel_run is one way [0]
* Under Linux, we do no ship with MKL, Gotoblas neither Atlas (but with
a very slow refblas). If you switch to one of them you will get both
great performances and all core being used during matrix computations.
* GPGPU: you can give a try to our GPGPU extensions (don't hesitate if
you have any feedback). [1]
* If you are more into grid/cloud computing, you can give a try to proactive
[2]
* In Scilab, even it is a bit outdated, you will find a PVM interface [3].
* Finally, if you are familiar on the build process of an linux application,
you can try with the MPI implementation of Scilab (in the git branch "MPI")

Hope this helps,

Sylvestre

[0] http://help.scilab.org/parallel_run
[1] http://atoms.scilab.org/toolboxes/sciGPGPU
[2] http://proactive.inria.fr/
[3] http://help.scilab.org/pvm




If you reply to this email, your message will be added to the discussion below:

http://mailinglists.scilab.org/Parallel-execution-tp3513234p3526453.html

To unsubscribe from Parallel execution, click here.
NAML

sylvestre sylvestre
Reply | Threaded
Open this post in threaded view
|  
Report Content as Inappropriate

Re: Parallel execution

Le lundi 21 novembre 2011 à 23:19 -0800, [hidden email] a
écrit :
> Dear Sylvestre !
>
>
> Thank you for the reply I'm using scilab on a SGI UV 1000 with 2000
> cores
>
> I try the example programm  with no params that's why i chrashed , yes
> ok, but how can i prevent the users to do the sane as i did .
Could you report a bug ?

>  Currently i try with limit maxproec 128  (not 16 000 as default
> parameter)  on the command line.I didnt test it right now becouse i
> have to be sure that nobody is on the machine exept me
> otherwise :-( my boss will cut me into pieces.
You should change your boss ;)

Sylvestre


christian.wohlschlager@jku.at christian.wohlschlager@jku.at
Reply | Threaded
Open this post in threaded view
|  
Report Content as Inappropriate

Re: Parallel execution

Ok i will send a bug report.




>>> "sylvestre [via Scilab / Xcos - Mailing Lists Archives]" <ml-node+[hidden email]> 11/22/2011 10:17 AM >>>
Le lundi 21 novembre 2011 Ã 23:19 -0800, [hidden email] a
écrit :
> Dear Sylvestre !
>
>
> Thank you for the reply I'm using scilab on a SGI UV 1000 with 2000
> cores
>
> I try the example programm  with no params that's why i chrashed , yes
> ok, but how can i prevent the users to do the sane as i did .
Could you report a bug ?

>  Currently i try with limit maxproec 128  (not 16 000 as default
> parameter)  on the command line.I didnt test it right now becouse i
> have to be sure that nobody is on the machine exept me
> otherwise :-( my boss will cut me into pieces.
You should change your boss ;)

Sylvestre





If you reply to this email, your message will be added to the discussion below:

http://mailinglists.scilab.org/Parallel-execution-tp3513234p3527208.html

To unsubscribe from Parallel execution, click here.
NAML

Loading...