AMBER Archive (2004)

Subject: Re: AMBER: installing amber8 on a dual processor single box

From: Robert Duke (rduke_at_email.unc.edu)
Date: Fri Apr 02 2004 - 12:58:35 CST


Carsten -
Thanks for posting the summary. Three points:
1) The intel compiler referenced here probably needs at least redhat 7.3,
since 7.2 is glibc 2.2.4.
2) When installing mpich, it would be best to use the ch_shmem device only
if you are not going to use multiple boxes, connected with something like
gigabit ethernet; the most versatile driver overall is the ch_p4 device,
which will handle smp's and ethernet-connected boxes.
3) I strongly recommend that if you have anything from redhat 8.0 onward,
you be sure that it is "up2date", with the latest kernel and libraries
available for that release. This will hopefully help you avoid some of the
potential nightmares with threads code.
Regards - Bob

----- Original Message -----
From: "Carsten Detering" <detering_at_u.washington.edu>
To: <amber_at_scripps.edu>
Sent: Friday, April 02, 2004 1:37 PM
Subject: AMBER: installing amber8 on a dual processor single box

> Dear all,
>
> after a more or less troublesome set-up (thanks to Bob Duke from UNC at
this
> point), I would like to share the experience of setting up Amber8 on a
dual
> processor machine (Redhat 8) using MPICH-1.2.5.2 and the new Intel
fortran
> 90 compiler. I guess this works for Redhat releases down to 7.2. Correct
me
> if I am wrong.
>
> 1. Download the compiler from http://premier.intel.com. You have to create
> an account beforehand on www.intel.com; you will get password and login by
> email. THey said these would work for the premier site as well, but at the
> time I was doing it there was something wrong with their web site and I
was
> not able to log in. Contact quadsupport if it does not work.
> After having successfully logged in to the premier site, download the
> l_fc_pc_7[2].1.038.tar file. According to Bob Duke, this compiler supports
> glib2.2.5, 2.2.93, and 2.3.2 (see the mail reflector for his detailed
email
> to me about this). After you downloaded the file and untared, copy the
> license file you have gotten from intel by mail in the same directory as
the
> compiler (I don't know if that's really necessary for installation,
though).
> Then install the compiler. The location of the compiler should be
> /opt/intel/compiler70/ia32/bin, unless you specified something else. Make
> sure this is in your path.
> 2.Download and install MPICH. The latest version is 1.2.5.2. For
> installation, set the following environment variables (this is taken from
> Bob Dukes pmemd3.00 release notes with a little adjustment for single box
> dual processor machine - thanks)
> setenv FC ifc
> setenv F90 ifc
> setenv FLIB_LIST -lPEPCF90
> setenv FFLAGS "-auto -tpp6 -mp1" (works for Athlon)
> setenv RSHCOMMAND /usr/bin/rsh (depends on your installation)
> ./configure --disable-devdebug \
> -prefix=/install-directory-for-mpich -optf77=-O2 -optcc=
-O
> 2 \
> --with-device=ch_shmem \
> --disable-c++
> THis should work. If you had MPICH before and used it with amber7, it
> depends which compiler you compiled it with. Compilation of AMber8 in
> parallel would not work if you had compiled MPICH with a g77. In case of
> doubt re-install MPICH clean (this is what I did), and recompile it using
> the settings above.
> 3. Go to the /amber8/src directory. You should have installed amber8 in
> serial mode first, according to the manual (I used './configure -p4 ifc(or
> ifc7)' and then 'make serial'). THis worked flawlessly for me. Possible
> failures were in marginal rounding errors. Before compiling for parallel,
> make sure you have set the MPICH_HOME dir to the MPICH-installation-dir,
and
> the DO_PARALLEL to 'mpirun -np 2', if you have two processors. THen do as
> described in the manual: make clean, ./configure -mpich ifc, make
parallel.
> Now for the test, just type 'make test' ('make test.parallel' does not
> work). Look up the *.dif files if you get possible failures, the error
> should be in the third digit after the komma.
>
> I hope this will be useful for some people.
>
>
> Cheers,
> Carsten
>
>
> --------------------------------------------
> Carsten Detering
> University of Washington
> Department of Chemistry
> Seattle, WA 98195
> Fon 206.543.5081
> Fax 206.685.8665
>
> -----------------------------------------------------------------------
> The AMBER Mail Reflector
> To post, send mail to amber_at_scripps.edu
> To unsubscribe, send "unsubscribe amber" to majordomo_at_scripps.edu
>

-----------------------------------------------------------------------
The AMBER Mail Reflector
To post, send mail to amber_at_scripps.edu
To unsubscribe, send "unsubscribe amber" to majordomo_at_scripps.edu