×

Notice

The forum is in read only mode.

Code_saturne 1.4b trouble [install]

More
16 years 5 months ago #2371 by Claus
Replied by Claus on topic Re:Code_saturne 1.4b trouble [install]
Im pretty sure I use the scotch wrapper for metis - thats the only thing I found in the package-manager. (can you confirm this from the output?)

claus@claus-desktop:~$ ldd /opt/code_saturne/Enveloppe/ecs-1.4.a/arch/Linux/bin/cs_partition
linux-gate.so.1 => (0xb7f0d000)
libm.so.6 => /lib/tls/i686/cmov/libm.so.6 (0x4aa8a000)
libbft.so.1 => /opt/code_saturne/opt/bft-1.0.7/arch/Linux/lib/libbft.so.1 (0xb7ee7000)
libz.so.1 => /usr/lib/libz.so.1 (0x4aacd000)
libc.so.6 => /lib/tls/i686/cmov/libc.so.6 (0x4a924000)
/lib/ld-linux.so.2 (0x44595000)


So please remind me; If I want to install Metis-4 during compile, I should extract metis to $cs_root/src/metis-4 and enter the path in 'setup' and answer 'yes' to use and 'yes' to install?

EDIT: I didn't read your answer well enough, I'll try to configure metis now :pinch: :blush:<br /><br />Post edited by: Claus, at: 2008/11/12 15:47

Code_Aster release : STA11.4 on OpenSUSE 12.3 64 bits - EDF/Intel version
More
16 years 5 months ago #2372 by Claus
Replied by Claus on topic Re:Code_saturne 1.4b trouble [install]
I tried with Metis-4.0 now, but still get an error on the partitioner:

claus@claus-desktop:~/Stuff/CAD/STRATIFIED_JUNCTION/NEW/DATA$ ldd /opt/code_saturne/Enveloppe/ecs-1.4.a/arch/Linux/bin/cs_partition
linux-gate.so.1 =&gt; (0xb8076000)
libm.so.6 =&gt; /lib/tls/i686/cmov/libm.so.6 (0x4aa8a000)
libbft.so.1 =&gt; /opt/code_saturne/opt/bft-1.0.7/arch/Linux/lib/libbft.so.1 (0xb8050000)
libz.so.1 =&gt; /usr/lib/libz.so.1 (0x4aacd000)
libc.so.6 =&gt; /lib/tls/i686/cmov/libc.so.6 (0x4a924000)
/lib/ld-linux.so.2 (0x44595000)


Heres my setup file from the Install folder:

#
# Architecture Name
#
#nom_arch Linux &lt;- for specific architecture
#nom_arch &lt;- leave empty for automatic default
# value (based on uname)
nom_arch Linux
#
#
# C compiler
#
compC gcc
#
#
# Fortran compiler
#
compF gfortran
#
#
# Language
# default: &quot;EN&quot; english
# others: &quot;FR&quot; french
#
language EN
#
#
# Name Path LibName Use Install
#
#
#
# Python is mandatory to launch the Graphical User
# Interface. It has to be compiled with Tcl/Tk/Tix
# support.
#
# It is highly recommended to use the Python provided
# by the distribution and to install Tcl/Tk/Tix through
# the package manager if needed. The user can test if
# the installation process has been correctly done by
# typing in a Python prompt &quot;import Tix&quot;
#
# The following setup line will usually not be changed
# except if the user has installed his proper Python.
#
#
python /usr python yes no
#
#
# This package allows for hardware-optimized BLAS
# (Basic Linear Algebra Subroutines). If no system
# BLAS is used, we revert to an internal BLAS
# emulation, which may be somewhat slower.
#
# ATLAS (or another BLAS) should be available for most
# platforms through the package manager.
#
# Otherwise, only ATLAS installation on 32 bit systems
# is currently handled by this script.
#
# If using the Intel or IBM compilers, IMKL or ESSL
# may be used in place of ATLAS respectively.
# If so, Code_Saturne Kernel installation must be
# done manually.
#
#
blas /usr atlas-3.8.2 yes no
#
#
# Optional packages:
#
# MED / HDF5 For MED file format support
# (used by Salome and now by Gmsh)
#
# CGNS For CGNS file support
# (used by many meshers)
#
# Open MPI (or MPICH2) + Metis
#
# For massively parallel architectures, it is
# recommended to use the system's default
# MPI library.
#
# If Open MPI is not chosen as the MPI
# distribution for the code, Code_Saturne
# Kernel has to be installed manually.
#
# For Linux workstations, MPI, HDF5, and even
# MED packages may be available through the
# package manager. HDF5 is also often
# available on large systems such as
# IBM Blue Gene or Cray XT.
#
# Metis is more rarely found in Linux
# distributions, but may already be
# installed on massively parallel machines
# and on clusters. For good parallel
# performance, it is highly recommended.
# For meshes larger than 15 million cells,
# Metis 5.0 beta is recommended, as Metis 4
# may fail above the 20-35 million cells.
#
# If Metis 5.0 is chosen, its installation
# must be done manually.
#
#
metis /opt/code_saturne/opt/metis-4.0 metis-4.0 yes no
mpi /usr/lib/openmpi openmpi-1.2.7 yes no
cgns /usr cgns-2.5.3 yes no
hdf /usr hdf5-1.6.7 yes no
med /usr med-2.3.4 yes no
#
#
# Code_Saturne kernel and module libraries
#
#
bft /opt/code_saturne/opt/bft-1.0.7/arch/Linux bft-1.0.7 yes no
fvm /opt/code_saturne/opt/fvm-0.11.0/arch/Linux fvm-0.11.0 yes no
ecs /opt/code_saturne/Enveloppe/ecs-1.4.a/arch/Linux ecs-1.4.a yes no
ics /opt/code_saturne/Interface/ics-1.4.a ics-1.4.a yes no
ncs /opt/code_saturne/Noyau/ncs-1.4.b/arch/Linux ncs-1.4.b yes no
#
#
# Optional coupling with the SYRTHES thermal code
# Only the installation of syr_cs is proposed.
#
#
syr_cs None syr_cs-2.1.0 no no
syrthes None syrthes3.4.1 no no


This is the 'listpart'

.
.
| Code_Saturne Partitioner |
`
'

External library: METIS 4.0

Opening input file: &quot;preprocessor_output&quot;

Ignored: &quot;start_block:dimensions &quot;
Read: &quot;n_cells &quot;; Type: &quot;u8&quot;; Size: 1
Read: &quot;n_faces &quot;; Type: &quot;u8&quot;; Size: 1
Ignored: &quot;n_vertices &quot;; Type: &quot;u8&quot;; Size: 1
Ignored: &quot;face_vertices_size &quot;; Type: &quot;u8&quot;; Size: 1
Ignored: &quot;n_group_classes &quot;; Type: &quot;i4&quot;; Size: 1
Ignored: &quot;n_group_class_props_max &quot;; Type: &quot;i4&quot;; Size: 1
Ignored: &quot;group_class_properties &quot;; Type: &quot;i4&quot;; Size: 8
Ignored: &quot;end_block:dimensions &quot;
Ignored: &quot;start_block:data &quot;
Read: &quot;face_cells &quot;; Type: &quot;i4&quot;; Size: 101216
Ignored: &quot;cell_group_class_id &quot;; Type: &quot;i4&quot;; Size: 16320
Ignored: &quot;face_group_class_id &quot;; Type: &quot;i4&quot;; Size: 50608
Ignored: &quot;face_vertices_index &quot;; Type: &quot;u4&quot;; Size: 50609
Ignored: &quot;face_vertices &quot;; Type: &quot;i4&quot;; Size: 202432
Ignored: &quot;vertex_coords &quot;; Type: &quot;r8&quot;; Size: 54210
Ignored: &quot;end_block:data &quot;
Reached: &quot;EOF &quot;

Closing input: &quot;preprocessor_output&quot;

Partitioning 16320 cells to 2 domains (METIS_PartGraphRecursive).

Total number of faces on parallel boundaries: 203
wall-clock time: 0.030170 s; CPU time: 0.030000 s

Number of cells per domain (histogramm):
[ 0 ; 807 [ = 1
[ 807 ; 1615 [ = 0
[ 1615 ; 2423 [ = 0
[ 2423 ; 3231 [ = 0
[ 3231 ; 4038 [ = 0
[ 4038 ; 4846 [ = 0
[ 4846 ; 5654 [ = 0
[ 5654 ; 6462 [ = 0
[ 6462 ; 7270 [ = 0
[ 7270 ; 8078 ] = 1


Im unsure of what to do next. :huh:

Code_Aster release : STA11.4 on OpenSUSE 12.3 64 bits - EDF/Intel version
More
16 years 5 months ago #2373 by David Monfort
Replied by David Monfort on topic Re:Code_saturne 1.4b trouble [install]
I cannot tell you whether you are using SCOTCH, METIS or nothing from this output, but that doesn't mean much... It may be linked statically within cs_partition. I could answer you for sure (I think...) if you still have the file named config.log which is somewhere in the ecs-1.4.a directory, generally at the root of this directory.

As for the METIS library, if you are downloading the 4.0 version it should be fine to install it via the setup script (on the contrary, the 5.0pre2 cannot been installed via this script). Don't forget to re-install the Preprocessor as well! :P
More
16 years 5 months ago #2374 by David Monfort
Replied by David Monfort on topic Re:Code_saturne 1.4b trouble [install]
You also have to re-install the Preprocessor so that it uses METIS. Just to check on my computer, could you send me your mesh ? It doesn't seem too big.
More
16 years 5 months ago #2375 by Claus
Replied by Claus on topic Re:Code_saturne 1.4b trouble [install]
It seems to be using Metis-4.0 from the compil.log (attached)

Maybe I should try to clear the entire installation, and begin again.

Attachment ecs.bz2 not found




edit: The mesh is the mesh from the 'Stratified Junction' test-case. I just run that case to test CS-1.4

edit2: I tried completely reinstalling CS-1.4 but the partitioner still fails with 2 cpus.<br /><br />Post edited by: Claus, at: 2008/11/12 16:54

Code_Aster release : STA11.4 on OpenSUSE 12.3 64 bits - EDF/Intel version
Attachments:
More
16 years 5 months ago #2376 by David Monfort
Replied by David Monfort on topic Re:Code_saturne 1.4b trouble [install]
I've tested the partitionning in two domains the same mesh (from the tutorial) and do not raise any error. Here is the output from the partitionner tool:

.

.
| Code_Saturne Partitioner |
`
'

External library: METIS 4.0

Opening input file: &quot;preprocessor_output&quot;

Ignored: &quot;start_block:dimensions &quot;
Read: &quot;n_cells &quot;; Type: &quot;u8&quot;; Size: 1
Read: &quot;n_faces &quot;; Type: &quot;u8&quot;; Size: 1
Ignored: &quot;n_vertices &quot;; Type: &quot;u8&quot;; Size: 1
Ignored: &quot;face_vertices_size &quot;; Type: &quot;u8&quot;; Size: 1
Ignored: &quot;n_group_classes &quot;; Type: &quot;i4&quot;; Size: 1
Ignored: &quot;n_group_class_props_max &quot;; Type: &quot;i4&quot;; Size: 1
Ignored: &quot;group_class_properties &quot;; Type: &quot;i4&quot;; Size: 8
Ignored: &quot;end_block:dimensions &quot;
Ignored: &quot;start_block:data &quot;
Read: &quot;face_cells &quot;; Type: &quot;i4&quot;; Size: 101216
Ignored: &quot;cell_group_class_id &quot;; Type: &quot;i4&quot;; Size: 16320
Ignored: &quot;face_group_class_id &quot;; Type: &quot;i4&quot;; Size: 50608
Ignored: &quot;face_vertices_index &quot;; Type: &quot;u4&quot;; Size: 50609
Ignored: &quot;face_vertices &quot;; Type: &quot;i4&quot;; Size: 202432
Ignored: &quot;vertex_coords &quot;; Type: &quot;r8&quot;; Size: 54210
Ignored: &quot;end_block:data &quot;
Reached: &quot;EOF &quot;

Closing input: &quot;preprocessor_output&quot;

Partitioning 16320 cells to 2 domains (METIS_PartGraphRecursive).

Total number of faces on parallel boundaries: 208
wall-clock time: 0.010504 s; CPU time: 0.012001 s

Number of cells per domain (histogramm):
[ 8160 ; 8160 ] = 2

Opening output: domain_number_2

Closing output: &quot;domain_number_2&quot;

Memory use and timing summary:

Total memory: 11.199 Mb

Wall-clock time: 0.037 s
CPU time: 0.016 s

.
.
| Partitioner finish |
`
'

As you can see, the mesh is correctly partitionned into two domains with 8160 cells each.

As for your previous installation, it is indeed compiled with the METIS wrapper from the SCOTCH library. You can check this by &quot;grepping&quot; the HAVE_SCOTCH_METIS variable.

When you're saying it still fails, you mean with the same error? Could you check in your new config.log from the Preprocessor installation that the above variable is set to 0 and that HAVE_METIS is set to 1? Or post it on the forum if you have difficulties to read it (it's not always easy...)

David<br /><br />Post edited by: David Monfort, at: 2008/11/12 18:26
Moderators: catux
Time to create page: 0.148 seconds
Powered by Kunena Forum