Newsgroups: comp.parallel.mpi
From: "Y. Hamadi" <hamadi@lirmm.fr>
Subject: Re: Memory leak on the SP2 with MPI
Organization: LIRMM - FRANCE
Date: Thu, 23 Jul 1998 15:31:25 +0200
Mime-Version: 1.0
Content-Type: multipart/alternative; boundary="------------36092CC759DFA2A3D3D83E36"
Message-ID: <35B73B2D.C12AD775@lirmm.fr>


--------------36092CC759DFA2A3D3D83E36
Content-Type: text/plain; charset=us-ascii
Content-Transfer-Encoding: 7bit

L'Excellent Jean Yves wrote:

> Wai Sun Don wrote:
> >
> > Does anyone have memory leak problem on the SP2 with MPI?
> >
> > I am running a 3D code on the SP2 with MPI (version unknown, I think
> > it is still MPI-1).  I have track the memory leak to the MPI call
> > MPI_Barrier.  The Error return code is 0 for all 16 processors.
> >
> > ...
>
>    I had something similar on the SP2 when the MPI
> implementation was unable to allocate memory for internal
> purpose in some routines (such as MPI_BARRIER). There does
> not seem to be any user control on this so the only solution
> I could find was to try to reduce arbitrarily memory usage
> at the application level.
>
>    I would also be interested if someone knows more on this.
>
> Regards,
> Jean-Yves


Personnaly i've previously reported memory problems with the sp2
supercomputer running IBM mpi. The problems was also an internal memory
leak. Moreover when i try to allocate more memory by using the
MPI_attach_buffer with buffer larger size, the leak was at the same level
as precedently.
If it can help IBM people to give the right answers to people in this
newsgroup (i'm not alone now..). I can add that my distributed program
does not have memory leak with MPICH v1.1 on a local network of 12
sparc-5 workstations...

hope this help
yh

--
 Youssef HAMADI
  LIRMM - UMR 5506 CNRS/Universite Montpellier II
  161, Rue Ada  34392 MONTPELLIER CEDEX 5  FRANCE
  Tel: (+33) 04 67 41 86 13  Fax: (+33) 04 67 41 85 00
  http://www.lirmm.fr/~hamadi



--------------36092CC759DFA2A3D3D83E36
Content-Type: text/html; charset=us-ascii
Content-Transfer-Encoding: 7bit

<HTML>
L'Excellent Jean Yves wrote:
<BLOCKQUOTE TYPE=CITE>Wai Sun Don wrote:
<BR>>
<BR>> Does anyone have memory leak problem on the SP2 with MPI?
<BR>>
<BR>> I am running a 3D code on the SP2 with MPI (version unknown, I think
<BR>> it is still MPI-1).&nbsp; I have track the memory leak to the MPI
call
<BR>> MPI_Barrier.&nbsp; The Error return code is 0 for all 16 processors.
<BR>>
<BR>> ...

<P>&nbsp;&nbsp; I had something similar on the SP2 when the MPI
<BR>implementation was unable to allocate memory for internal
<BR>purpose in some routines (such as MPI_BARRIER). There does
<BR>not seem to be any user control on this so the only solution
<BR>I could find was to try to reduce arbitrarily memory usage
<BR>at the application level.

<P>&nbsp;&nbsp; I would also be interested if someone knows more on this.

<P>Regards,
<BR>Jean-Yves</BLOCKQUOTE>
&nbsp;
<BR>Personnaly i've previously reported memory problems with the sp2 supercomputer
running IBM mpi. The problems was also an internal memory leak. Moreover
when i try to allocate more memory by using the MPI_attach_buffer with
buffer larger size, the leak was at the same level as precedently.
<BR>If it can help IBM people to give the right answers to people in this
newsgroup (i'm not alone now..). I can add that my distributed program
does not have memory leak with MPICH v1.1 on a local network of 12 sparc-5
workstations...

<P>hope this help
<BR>yh
<PRE>--&nbsp;
&nbsp;Youssef HAMADI&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;
&nbsp; LIRMM - UMR 5506 CNRS/Universite Montpellier II&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;
&nbsp; 161, Rue Ada&nbsp; 34392 MONTPELLIER CEDEX 5&nbsp; FRANCE&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;
&nbsp; Tel: (+33) 04 67 41 86 13&nbsp; Fax: (+33) 04 67 41 85 00
&nbsp; <A HREF="http://www.lirmm.fr/~hamadi">http://www.lirmm.fr/~hamadi</A></PRE>
&nbsp;</HTML>

--------------36092CC759DFA2A3D3D83E36--


