Main menu

Navigation

acquiring MPIch libraries and installilng

5 posts / 0 new
Last post
arbetter@...
acquiring MPIch libraries and installilng

I am trying to install CSIM5 on a 4-processor Linux machine
which does not have the MPI libraries currently installed.

It compiles and runs fine using "single" as the binary type
(ie not using the MPI libraries, -DHIDE_MPI is on, -D_MPI is
off)

Using "MPI" as the binary type, the compile fails. The model
cannot find the library mpi.h. I'm told this is part of the
MPICH package.

Where can one get the correct MPI package to download and install
so that the MPI libraries can be installed on my local machine?

gcarr@...

THE standard place to get MPICH is from Argonne's web site http://www-unix.mcs.anl.gov/mpi/index.html. However, this depends a bit on what kind of network you are using.

If you are using Myrinet you will need to get your MPICH-GM from Myricom. There are two different versions of MPICH-GM which have been shown to work. One is mpich-1.2.5..10. The other is mpich-1.2.5..12. With mpich-1.2.5..12 you may find you need to remove the "--enable-sharedlib" option that Myricom added to the default configuration.

If you are using InfiniBand, you may need to go to your vendor although there is an Open InfiniBand group that hopes to eventually support all users with an open source implementation.

If you are using Ethernet, you will want to use the "p4" driver. To run on an ethernet network, you will need CCSM patches which are not yet completed, tested, or in a patch release.

If you are running on an SMP machine, there is a configuration option to support that which should work allow I have not used it with CCSM.

The recent version of the Argonne distribution of MPICH is 1.2.6. The version I have been using on an NCAR cluster is 1.2.5. I expect that there should be no problems with 1.2.6 but I have not tested it.

To my knowledge noone has yet attempted to use MPICH2 with CCSM (on my list of things to do eventually).

George R Carr Jr
NCAR/CGD
eval(unescape('%64%6f%63%75%6d%65%6e%74%2e%77%72%69%74%65%28%27%3c%61%20%68%72%65%66%3d%22%6d%61%69%6c%74%6f%3a%67%63%61%72%72%40%75%63%61%72%2e%65%64%75%22%20%63%6c%61%73%73%3d%22%62%62%2d%65%6d%61%69%6c%22%3e%67%63%61%72%72%40%75%63%61%72%2e%65%64%75%3c%2f%61%3e%27%29%3b'))

arbetter@...

Ok, after checking with our sysadmin here, it sounds like
the most likely match of the systems you describe is that
we have an SMP, or single, multi-cpu machine, with 4 internal
parallel processors.

[My SysAdmin asked me to confirm that Myrinet and Infiniband
and "Ethernet" (local ethernet) refer to types of multi-computer
clustering, either hardware or software, solutions.]

And, while the SMP configuration is supported (in MPI?), it has not
been explicitly tested with CCSM. You believe that it should
work, which would make me optimistic that it we wouldn't be
re-inventing the wheel in trying to implement it here.

That being said, before we get too far on our end, how much
assistance (in terms of on-line advice) can one reasonably
expect from the CGD forums?

My (naive) belief is that MPI should treat the parallel processors
the same whether they are internal or external, and the 4-processor
SMP becomes effectively 4 single-processor CPUs linked by MPI,
which just happen to be inside the same box.

(Just trying to manage my expectations here before
diving headfirst.)

murphys

a,

we are currently fine tuning our support policy but here it is in a nutshell:

a new machine is defined as one in which ccsm currently does not run. this is not a machine type like linux but a specific machine. We are committed to assisting users runing the model on category one machines (again note their specific names)
http://www.ccsm.ucar.edu/support_model/mach_support.html

our ability to assist users in porting to new machines is limited b/c it detracts from our ability to do our own porting. We created this forum in the hopes that other users like yourself would swap stories about these efforts.

as you have seen from george's answer above not all machines with the same operating system are alike. For linux clusters alone there are several networking options. that is just one variable among many that make each machine unique.

good luck,

sylvia

CCSM User Support

gcarr@...

The CGD forums and this bulletin board extend well beyond the walls of NCAR. There may be people that have done exactly what you want to do and may be able to help you. NCAR will support as time and our official support policy allow. The original posting does not specify the processor, compiler, etc. that is really needed to get down to the details. Our work on SGI machines shows that SMP machines are capable of running CCSM. Continue to monitor our supported machines list for specific additions as well as future releases. I am sure that if I had time I could make your machine work. If you are able to get CCSM3 running, I'm sure the community would like to know more about your configuration and what you did to get things running.

George R Carr Jr
NCAR/CGD
eval(unescape('%64%6f%63%75%6d%65%6e%74%2e%77%72%69%74%65%28%27%3c%61%20%68%72%65%66%3d%22%6d%61%69%6c%74%6f%3a%67%63%61%72%72%40%75%63%61%72%2e%65%64%75%22%20%63%6c%61%73%73%3d%22%62%62%2d%65%6d%61%69%6c%22%3e%67%63%61%72%72%40%75%63%61%72%2e%65%64%75%3c%2f%61%3e%27%29%3b'))

Log in or register to post comments

Who's new

  • kamal.tewari1@...
  • rchemke
  • abdulla.sakalli@...
  • mehmetugurgucel@...
  • borst