[CIG-LONG] CIG-LONG Digest, Vol 46, Issue 7
Laetitia Le Pourhiet
zlolita at gmail.com
Thu Sep 30 00:29:03 PDT 2010
and I forgot to mention, beacause I didn't see the email of John.
you should continue to use openmpi on your machine
because mpich doesn't deal very well with shared memory architecture.
it doesn't affect the job to one core only unless you create a fake
machine file and launch mpd which I found very complex. As a result,
using mpich on your machine will run the job on seven cpu if you ask for
8 and you will loose performance.
there is an option in openmpi to affect each process to a single core
and therefore override ubuntu which always try to save one core for
itself (the little B.)
so check your CPU usage and memory Usage by profiling your performance
if you encounter the problem, I will have more time on Friday to look
for the option again...
once again good luck... it is not easy to make mpi run efficiently on
the shared memory architectures...
Laetitia
Nicolas RIEL wrote:
> Thank you for your answers !
>
> I'm using Ubuntu, and I installed PETSc 3.0.0, however unlike you,
> John it's king of tricky !
> When, like you say Ali, i'm trying to install all the libraries from
> PETSc, everything is working well exept for blacs.
> I'm using this code :
> ./config/configure.py --with-fc90=gfortran --download-parmetis
> --download-mumps --download-blacs --download-scalapack
> --with-mpi-dir=/usr/bin/
> and the error is :
> Downloaded blacs could not be used. Please check install in
> /home/seph/petsc-3.0.0-p12/linux-gnu-c-debug.
>
> Into the text file contained in linux-gnu-c-debug, there is one file
> about blacs named "blacs"
>
> SHELL = /bin/sh
> COMMLIB = MPI
> SENDIS = -DSndIsLocBlk
> WHATMPI = -DUseMpi2
> DEBUGLVL = -DBlacsDebugLvl=1
> BLACSdir = /home/seph/petsc-3.0.0-p12/externalpackages/blacs-dev
> BLACSLIB = /home/seph/petsc-3.0.0-p12/linux-gnu-c-debug/lib/libblacs.a
> MPILIB =
> SYSINC = -I/usr/bin/include -I/usr/include/mpich2
> BTLIBS = $(BLACSLIB) $(MPILIB)
> INTFACE = -DAdd_
> DEFS1 = -DSYSINC $(SYSINC) $(INTFACE) $(DEFBSTOP) $(DEFCOMBTOP)
> $(DEBUGLVL)
> BLACSDEFS = $(DEFS1) $(SENDIS) $(BUFF) $(TRANSCOMM) $(WHATMPI)
> $(SYSERRORS)
> F77 = mpif90
> F77FLAGS = -Wno-unused-variable -g $(SYSINC)
> F77LOADER = mpif90
> F77LOADFLAGS = -Wall -Wno-unused-variable -g
> CC = mpicc
> CCFLAGS = -Wwrite-strings -Wno-strict-aliasing -g3
> CCLOADER = mpicc
> CCLOADFLAGS = -Wall -Wwrite-strings -Wno-strict-aliasing -g3
> ARCH = /usr/bin/ar
> ARCHFLAGS = cr
> RANLIB = /usr/bin/ranlib
>
> When i'm looking this code, I realise that blacs is looking for mpich,
> and I use openmpi with this line "SYSINC = -I/usr/bin/include
> -I/usr/include/mpich2" ?
>
> I'm almost new on linux, I still don't understand my mistake.
> But if I follow your advice, the best thing to do is to erase external
> libraries, and download and install everything through PETSc ?
>
>
>
> On Wed, Sep 29, 2010 at 9:00 PM, <cig-long-request at geodynamics.org
> <mailto:cig-long-request at geodynamics.org>> wrote:
>
> Send CIG-LONG mailing list submissions to
> cig-long at geodynamics.org <mailto:cig-long at geodynamics.org>
>
> To subscribe or unsubscribe via the World Wide Web, visit
> http://geodynamics.org/cgi-bin/mailman/listinfo/cig-long
> or, via email, send a message with subject or body 'help' to
> cig-long-request at geodynamics.org
> <mailto:cig-long-request at geodynamics.org>
>
> You can reach the person managing the list at
> cig-long-owner at geodynamics.org
> <mailto:cig-long-owner at geodynamics.org>
>
> When replying, please edit your Subject line so it is more specific
> than "Re: Contents of CIG-LONG digest..."
>
>
> Today's Topics:
>
> 1. Re: CIG-LONG Digest, Vol 46, Issue 3 (Nicolas RIEL)
> 2. Re: CIG-LONG Digest, Vol 46, Issue 3 (John Naliboff)
> 3. Re: CIG-LONG Digest, Vol 46, Issue 3 (Tabrez Ali)
>
>
> ----------------------------------------------------------------------
>
> Message: 1
> Date: Wed, 29 Sep 2010 18:25:07 +0200
> From: Nicolas RIEL <nicolas.riel at free.fr
> <mailto:nicolas.riel at free.fr>>
> Subject: Re: [CIG-LONG] CIG-LONG Digest, Vol 46, Issue 3
> To: cig-long at geodynamics.org <mailto:cig-long at geodynamics.org>
> Message-ID:
>
> <AANLkTim9TYJOA0uTwG718jjoC-N_jSC9jSJtm1+DSk7P at mail.gmail.com
> <mailto:AANLkTim9TYJOA0uTwG718jjoC-N_jSC9jSJtm1%2BDSk7P at mail.gmail.com>>
> Content-Type: text/plain; charset="iso-8859-1"
>
> Hello Everyone,
>
> I'm actually runing Gale in parallel on my personal computer
> (Processor
> intel i7 - with 8 cores, on linux), and it works fine
>
> However I tried to install "Mumps" for direct solving in parallel
> (Through
> reconfiguring of PETSc) and I'm encountering problems.
> Mumps needs some more libraries to work (Scalapack, which needs
> Blacs Lapack
> and blas)
> And I just can't install mumps which crashes when trying to
> compile "blacs".
> I tried to install Scalapack (which uses blacs library)
> independently, but
> It does not work better ! The problem is coming from "blacs", and
> I don't
> see how to resolve it.
> Any ideas ?
> -------------- next part --------------
> An HTML attachment was scrubbed...
> URL:
> http://geodynamics.org/pipermail/cig-long/attachments/20100929/9d5733ae/attachment.html
>
> ------------------------------
>
> Message: 2
> Date: Wed, 29 Sep 2010 10:07:54 -0700
> From: John Naliboff <jbnaliboff at ucdavis.edu
> <mailto:jbnaliboff at ucdavis.edu>>
> Subject: Re: [CIG-LONG] CIG-LONG Digest, Vol 46, Issue 3
> To: cig-long at geodynamics.org <mailto:cig-long at geodynamics.org>
> Message-ID: <4302FE6F-DDA3-4648-95EA-F9DF8E48A84F at ucdavis.edu
> <mailto:4302FE6F-DDA3-4648-95EA-F9DF8E48A84F at ucdavis.edu>>
> Content-Type: text/plain; charset=us-ascii
>
> Hi Nicolas,
>
> I also had trouble installing Mumps when I was using PETSc 2.3.2.
> The fix was relatively minor, but I did not encounter the issue
> when installing Mumps with PETSc 3.0. If I remember correctly,
> the error with Mumps and PETSc 2.3.2 was also a Mac OS X specific
> issue.
>
> What version of PETSc are you using? If it is PETSc 2.3.2, I'll
> send you over the modification that allowed for a successful
> installation of Mumps, but that fix may very well be independent
> of your issue.
>
> Also, the PETSc folks were able to provide the fix for my
> installation problem after a few emails of discussion.
>
> Cheers,
> John
>
>
>
> On Sep 29, 2010, at 9:25 AM, Nicolas RIEL wrote:
>
> > Hello Everyone,
> >
> > I'm actually runing Gale in parallel on my personal computer
> (Processor intel i7 - with 8 cores, on linux), and it works fine
> >
> > However I tried to install "Mumps" for direct solving in
> parallel (Through reconfiguring of PETSc) and I'm encountering
> problems.
> > Mumps needs some more libraries to work (Scalapack, which needs
> Blacs Lapack and blas)
> > And I just can't install mumps which crashes when trying to
> compile "blacs".
> > I tried to install Scalapack (which uses blacs library)
> independently, but It does not work better ! The problem is coming
> from "blacs", and I don't see how to resolve it.
> > Any ideas ?
> >
> > _______________________________________________
> > CIG-LONG mailing list
> > CIG-LONG at geodynamics.org <mailto:CIG-LONG at geodynamics.org>
> > http://geodynamics.org/cgi-bin/mailman/listinfo/cig-long
>
>
>
> ------------------------------
>
> Message: 3
> Date: Wed, 29 Sep 2010 12:09:03 -0500
> From: Tabrez Ali <stali at geology.wisc.edu
> <mailto:stali at geology.wisc.edu>>
> Subject: Re: [CIG-LONG] CIG-LONG Digest, Vol 46, Issue 3
> To: Nicolas RIEL <nicolas.riel at free.fr <mailto:nicolas.riel at free.fr>>
> Cc: cig-long at geodynamics.org <mailto:cig-long at geodynamics.org>
> Message-ID: <24F25C10-03B4-47BD-9DA3-429050B67081 at geology.wisc.edu
> <mailto:24F25C10-03B4-47BD-9DA3-429050B67081 at geology.wisc.edu>>
> Content-Type: text/plain; charset=US-ASCII; format=flowed; delsp=yes
>
> I have some old notes which may be helpful.
>
> - http://www.geology.wisc.edu/~stali/old/parlibs.htm
> <http://www.geology.wisc.edu/%7Estali/old/parlibs.htm>
>
>
> However a very easy way is to install petsc and configure it so that
> it downloads and builds mpi, blas, lapack, blacs, scalapack, mumps etc
> for you. The built libs can be found in the "externalpackages" sub-
> directory within petsc.
>
> - Now you can use all these libs via petsc and also by itself as long
> as you link properly.
>
> - At this point if you dont want petsc and only want to keep mpi,
> blas, lapack etc. then simply move the contents of externalpackages
> directory to some place else (eg /opt), delete petsc and use the new
> path for linking
>
>
> To install petsc with above libs use something like:
>
> ./configure --with-mpi-dir=/opt/mpich2-1.0.8 --with-mumps=1
> --download-
> mumps=ifneeded --with-scalapack=1 --download-scalapack=ifneeded
> --with-
> blacs=1 --download-blacs=ifneeded --download-f-blas-lapack=ifneeded
> etc.
>
>
> On Sep 29, 2010, at 11:25 AM, Nicolas RIEL wrote:
>
> > Hello Everyone,
> >
> > I'm actually runing Gale in parallel on my personal computer
> > (Processor intel i7 - with 8 cores, on linux), and it works fine
> >
> > However I tried to install "Mumps" for direct solving in parallel
> > (Through reconfiguring of PETSc) and I'm encountering problems.
> > Mumps needs some more libraries to work (Scalapack, which needs
> > Blacs Lapack and blas)
> > And I just can't install mumps which crashes when trying to compile
> > "blacs".
> > I tried to install Scalapack (which uses blacs library)
> > independently, but It does not work better ! The problem is coming
> > from "blacs", and I don't see how to resolve it.
> > Any ideas ?
> >
> > _______________________________________________
> > CIG-LONG mailing list
> > CIG-LONG at geodynamics.org <mailto:CIG-LONG at geodynamics.org>
> > http://geodynamics.org/cgi-bin/mailman/listinfo/cig-long
>
>
>
> ------------------------------
>
> _______________________________________________
> CIG-LONG mailing list
> CIG-LONG at geodynamics.org <mailto:CIG-LONG at geodynamics.org>
> http://geodynamics.org/cgi-bin/mailman/listinfo/cig-long
>
>
> End of CIG-LONG Digest, Vol 46, Issue 7
> ***************************************
>
>
> ------------------------------------------------------------------------
>
> _______________________________________________
> CIG-LONG mailing list
> CIG-LONG at geodynamics.org
> http://geodynamics.org/cgi-bin/mailman/listinfo/cig-long
>
-------------- next part --------------
A non-text attachment was scrubbed...
Name: zlolita.vcf
Type: text/x-vcard
Size: 321 bytes
Desc: not available
Url : http://geodynamics.org/pipermail/cig-long/attachments/20100930/9e7c096b/attachment-0001.vcf
More information about the CIG-LONG
mailing list