|
 Need advice for compilation (to use in parallel) on a SGI Altix 4700
|
Joined: May 2007
Posts: 3
Forum Member
|
OP
Forum Member
Joined: May 2007
Posts: 3 |
Dear CHARMM community,
We have managed to compile charmm (c33b2) on an SGI Altix 4700 (itanium2) system using:
./install.com gnu huge efc
(We tried first using either 'altix' or 'itanium' as $chm_host but the results were inferior, with many tests giving segmentation faults)
We tweaked a bit with the compilers flags (essentially changing '-tpp2' to '-mtune=itanium2 -mcpu=itanium2') This compilation uses icc/ifort (v. 9.0; we had linking problems with version 10) Such build passes all the tests, except for four that end abnormally:
<** c28test : mkpres **> <** c31test : swm4_liquid **> <** c33test : hqbm_single **> <** c33test : nawat **>
There are no segmentation faults. The 331 tests run in 28min 9s using:
CPU time : 1685.41 sec. Max Memory : 449 MB Max Swap : 4130 MB
Max Processes : 5 Max Threads : 4
This is not too bad (I believe), but the problem is that we don't find the right combination for using MPI libraries. We have tried both mpich2 and lam/mpi. Equally unsuccessfully. If you have any ideas about it, we will appreciate a lot hearing about them!
Thank you very much in advance for your help.
Cheers,
Miguel
Last edited by MOL; 06/15/07 10:45 AM.
|
|
|
 Re: Need advice for compilation (to use in parallel) on a SGI Altix 47
|
Joined: Sep 2003
Posts: 8,522 Likes: 2
Forum Member
|
Forum Member
Joined: Sep 2003
Posts: 8,522 Likes: 2 |
In theory either should work, although I think MPICH-1 usage has been more prevalent. Note that the MPI lib of choice should generally be compiled and configured with the same C and Fortran compilers as will be used to compile CHARMM.
Doesn't SGI supply an MPI lib for Altix systems?
|
|
|
 Re: Need advice for compilation (to use in parallel) on a SGI Altix 47
|
Joined: May 2007
Posts: 3
Forum Member
|
OP
Forum Member
Joined: May 2007
Posts: 3 |
Hi Rick,
Thank you.
SGI ships their Altix with the Intel MPI libraries. Apparently, there are issues with other software in the cluster, so we're using either LAM/MPI or MPICH2. We also have intel compilers (icc and ifort) that we have used to compile the MPI libraries. One problem is that MPICH2 doesn't support shared libraries for the intel compilers, and I think this is calling all sort of trouble when compiling CHARMM.
We will keep trying with MPICH2, then. Perhaps using gcc/gfortran. I will post here if we succeed.
If anyone has advice to give concerning compilers/libraries versions/combination, it will be very much appreciated.
Cheers,
Miguel
|
|
|
 Re: Need advice for compilation (to use in parallel) on a SGI Altix 47
|
Joined: Feb 2004
Posts: 147
Forum Member
|
Forum Member
Joined: Feb 2004
Posts: 147 |
To take advantage of the high-speed interconnect of Altix you should use the MPI libraries that are delivered by SGI. I really wonder how you configure LAM/MPI and MPICH2 for this machine...
To receive more help from this forum you should give more details than "we don't find the right combination for using MPI libraries".
|
|
|
 Re: Need advice for compilation (to use in parallel) on a SGI Altix 47
|
Joined: May 2007
Posts: 3
Forum Member
|
OP
Forum Member
Joined: May 2007
Posts: 3 |
Thank you. I thought I had given enough info. My question was regarding to the versions of compilers / mpi libraries.
Indeed, I have compiled already charmm using mpich2 (intel compilers version 9.0, mpich2 1.0.5) It kind of works, whit a speed improvement of nearly 3x under certain conditions.
But I was waiting before posting here what I did because following Rick's advice, that you reproduce, I also compiled charmm using the SGI MPI libraries (in fact, Intel MPI) The compilation was fine, no errors. But I haven't been able to test the binary, because there was an issue with the module (arsess) that is required to launch array sessions in the Altix. Our system manager has been on holidays till next Monday. Hopefully, we'll deal with the problem and then I'll post here the results and possibly new, more especific, questions.
Best,
Miguel
|
|
|
|
|
|