Dear all:
About two weeks ago, I published my amber installation problems here.
Now, they have been resolved and I put our solution here, you can refer
to it if you have the same problems.
Actually, it is not because of the MPI, the machine or something else,
the reason is the file transfer mode. We only changed the file transfer
mode from Auto select to ASCII and then everything was OK. When we
upload or download files between our PCs and the clusters, we have to
use the SSH Secure File Transfer tool (or some other tools), the
Operation/File Transfer Mode manual item can help us to select the mode
we need. The default option is "Auto select" and in most of the
transfers I know, it is no need to change it, except for Gaussian (a
classic Quantum Chemistry application) input files. I don't know why.
PS. If we upload a Gaussian input file written in txt format with an
Auto Select mode, then in Linux, at the end of each line, there is a
strange symbol "^M" in light grey.
The following is two former letters I issued:
Dear all:
We are trying to install Amber8 on our PC-cluster with athlon platform,
MPICH and compiler ifort and have finished the compiling works; however,
when testing parallel works, something very strange happened. I issue
the steps of compiling and problems in my test works here:
Setenv AMBERHOME /user/local/amber8
Setenv MPICH_HOME /user/local.....
../configure -athlon -static ifort
Make serial
Make test.<programme-name>
Make clean
../configure -mpich -static ifort
Make parallel
At the end of the compiling, the machine told me everything was ok and
the single threaded works passed the test smoothly. But when I did
these:
Cd /AMBERHOME/test
Setenv DO_PARALLEL "mpirun -np 4" (or "/user/local/mpich_ifc/bin/mpirun
-np 4)
Make test.parallel
The machine told me like this:
[nnizb.nemo test]$ make test.parallel
make: *** No rule to make target `test.parallel'. Stop.
When I test one of my works, I issued command like these:
[nnizb.nemo test]$ mpirun -np 4 sander -O -i 888.in -p 888.parm -c
888.crd -o 888.out -r 888.rst
or
[nnizb.nemo test]$ /user/local/mpich_ifc/bin/mpirun -np 4 sander -O -i
888.in -p 888.parm -c 888.crd -o 888.out -r 888.rst
And the error messages were given like:
[0] MPI Abort by user Aborting program !
[0] Aborting program!
p0_12768: p4_error: : 1
Killed by signal 2.
Killed by signal 2.
Killed by signal 2.
Our OS is npaci rocks 3.3 linux (it is based on Redhat linux and very
very similar with Redhat linux).
We also failed when we tried to install amber8 on another cluster. The
administrator told us that was due to the lack of MPI library on each
node, so he has to install the library on each node. Is this true? Do
you need to do this before you install amber on your machine?
Thank you!
Zhang Bing
National University of Singapore
Nanoscience and Nanotechnology Initiative
nnizb.nus.edu.sg
Dear all:
I got the test jobs passed by issuing the new test command. However,
when I tried my own works, it failed again:
[nnizb.nemo exe]$ mpirun -np 4 sander -O -i min.in -p cdk5.parm -c
cdk5.crd -o cdk.out -r cdk.rst
[0] MPI Abort by user Aborting program !
[0] Aborting program!
p0_3517: p4_error: : 1
Killed by signal 2.
Killed by signal 2.
Killed by signal 2.
I think it is something wrong with my command line. 'Coz I did see 4
processors running there simultaneously after I submitted test jobs by
"make test.sander". So now, I am wondering how you made your script for
the parallel sander and why you need a DO_PARALLEL environment variable
first. I tried to set such a variable before my "mpirun -np 4 sander -O
-i min.in -p cdk5.parm -c cdk5.crd -o cdk.out -r cdk.rst" command, it
did not work. So would you kindly show me the script or the command line
for a parallel sander? Any advice from you will be greatly appreciated.
Sincerely yours,
Zhang Bing
NUSNNI
-----------------------------------------------------------------------
The AMBER Mail Reflector
To post, send mail to amber.scripps.edu
To unsubscribe, send "unsubscribe amber" to majordomo.scripps.edu
Received on Wed Jun 15 2005 - 03:53:00 PDT
