AMBER Archive (2009)

Subject: Re: [AMBER] question about failures

From: peker milas (pekermilas_at_gmail.com)
Date: Tue Dec 08 2009 - 00:25:10 CST

Thanks again,

What i understood i probably won't need TIP5P solvent model. Which
suggests i may ignore it for now. About the parallelization issue, it
s not a personal computer (rather belong to my lab which was supposed
to work as server or workstation) but based on
my and my collegues' experience on debian and ubuntu we would like to
use ubuntu as our operating system. After your email i started to be
suspicious about it though... I have been faced with bash and sh
problem before, on another machine, so i just fixed them on this
machine too already. I installed all the related packages like
bison,flex,csh and so on, before installation of Amber. I also did
relevant changes on my .bashrc and .profile files such as setting up
AMBERHOME and MPI_HOME. I honestly can't tell anyting about the mpi
executables, because i didn't check them in detail and this can be the
only problematic point about parallel installation i believe. I will
definitely check them and i will remove all parallelization packages
then i will reinstall just one of them. I hope this will solve my
problem because otherwise i have no idea about what to do or how to
fix it...After all thank you so much for your help...

best
peker

On Mon, Dec 7, 2009 at 10:00 PM, Jason Swails <jason.swails_at_gmail.com> wrote:
> Hi,
>
> Parallel installations can be difficult to complete, especially on your own
> machine when you try to "fix" it and/or try out other mpi's.  TIP5P is an
> explicit solvent model that uses "extra points" on water, i.e. point charges
> on all three atom centers and two additional charges to help model some of
> water's behavior a little bit better.  The most common model is TIP3P
> generally, and TIP4P is less common, though still used.  TIP5P is, in my
> experience, used only for specific applications regarding some more complex
> behavior of water.  One thing I'd remind you to do is to download and apply
> all bugfixes from the amber website, as some of your issues may be handled
> in those.
>
> I'm going to assume this is your own machine since it's running Ubuntu which
> is far more common on a desktop platform than an HPC one.  As user-friendly
> as the distro is, it has its caveats.  You will want to install the
> following applications through aptitude (sudo apt-get install): csh, flex,
> bison/byacc (and obviously gfortran if you don't have it already).
> Furthermore, /bin/sh in Ubuntu systems points to /bin/dash rather than
> /bin/bash like most other distros, so you'll want to redirect that link
> (sudo rm /bin/sh; sudo ln /bin/bash /bin/sh).
>
> Regarding MPI and parallelization: you should use one.  Pick one and stick
> with it.  The last thing you want to do is wind up with a hodgepodge of
> different MPIs lying around your system in which its pieces are
> incompatible.  If you're installing the MPIs either through aptitude or
> building it from source using ./configure; make; sudo make install, it's
> going to put it in your global path (unless you tell it not to), so each MPI
> installation is going to overwrite the last (though maybe not
> completely...)  If you stick with one, you will not run into this problem.
> I personally use MPICH2 on my machine (Ubuntu, 4-core) and it has worked
> with all ambers/ambertools, though I've also gotten it to work with other
> MPIs (OpenMPI, mvapich) just as easily.
>
> Furthermore, make sure that you don't have /usr/bin mpi executables (and
> /usr/lib and /usr/include libraries and header files) in addition to files
> in /usr/local/(bin, lib, include) that you're really trying to use.  If
> you're still having problems installing amber10 in parallel, try again (be
> sure to "make clean" before you try a new installation though, as you need
> to start from scratch when applying different options).  You can search the
> mail reflector (or google) for your error messages to see if they've been
> discussed before (they likely have).  If they haven't you can post a new
> message with your error and you should get a response.
>
> (Also, when you're running the parallel tests, don't forget to set the
> DO_PARALLEL environment variable: export DO_PARALLEL="mpirun -np 4 ", or
> however many processors you want to devote to the test cases).
>
> Good luck!
> Jason
>
> On Tue, Dec 8, 2009 at 12:31 AM, peker milas <pekermilas_at_gmail.com> wrote:
>
>> Hi again,
>>
>> you have no idea how i m grateful for this quick reply, so thank you
>> again. What i understand from your email, this is an important
>> problem because i would like to use this program with all of its
>> features. Now, it looks like being more explicit about the situation
>> is needed. First of all this machine has 2 intel nehalem cpu s (8
>> physical cores and 16 logical cores) on it, with 8Gb ram. It also has
>> ubuntu 9.10 (kermic i believe) as its operating system. Its compiler
>> set is gcc 4.4.1. I installed AmberTools1.2 using gcc. There was some
>> bugs in configuration script and in fortran.cpp file which i fixed.
>> And then, probably more than 10 times i tried to install Amber10
>> parallel. While i was installing i believe i used all relevant
>> paralellizations (mpich,openmpi, even lam which was the default
>> parallelization method, so that comes with amber ). Every single trial
>> failed when i run tests. What i meant is the program started doing
>> either the first or the second test but it could never finish. This
>> was frustrating enough and i decided to install it first serially.
>> When i did that i got those errors. What i m trying to say here i
>> really lost in this installation process and i will definitely need
>> help for both types of installation. Again if somebody wants to help
>> it will be appreciated deeply...
>>
>> thank you so much...
>> peker
>>
>> On Mon, Dec 7, 2009 at 11:51 PM, Jason Swails <jason.swails_at_gmail.com>
>> wrote:
>> > All but one of these differences seem benign.  The tip5p test, however,
>> is
>> > slightly disconcerting since it doesn't appear that any output was
>> created
>> > by the test run comparable to that blip that's printed in
>> > TEST_FAILURES.dif.  If you plan on using that solvent model, I'd check
>> out
>> > that test case a little more carefully.
>> >
>> > Good luck!
>> > Jason
>> >
>> > On Mon, Dec 7, 2009 at 10:37 PM, peker milas <pekermilas_at_gmail.com>
>> wrote:
>> >
>> >> Dear Amber users,
>> >>
>> >> I recently installed AmberTools 1.2 and Amber 10 to a pc. Just for now
>> >> it is serial but i m planning to make it parallel in a very near
>> >> future. At the same time i had some test failures. I will attach my
>> >> test failures file to this email. My question is about those failures.
>> >> If anybody can explain me whether they are crucial or not also whether
>> >> i should fix them  i will be very happy.
>> >>
>> >> Thank you so much
>> >> peker
>> >>
>> >> _______________________________________________
>> >> AMBER mailing list
>> >> AMBER_at_ambermd.org
>> >> http://lists.ambermd.org/mailman/listinfo/amber
>> >>
>> >>
>> >
>> >
>> > --
>> > ---------------------------------------
>> > Jason M. Swails
>> > Quantum Theory Project,
>> > University of Florida
>> > Ph.D. Graduate Student
>> > 352-392-4032
>> > _______________________________________________
>> > AMBER mailing list
>> > AMBER_at_ambermd.org
>> > http://lists.ambermd.org/mailman/listinfo/amber
>> >
>>
>> _______________________________________________
>> AMBER mailing list
>> AMBER_at_ambermd.org
>> http://lists.ambermd.org/mailman/listinfo/amber
>>
>
>
>
> --
> ---------------------------------------
> Jason M. Swails
> Quantum Theory Project,
> University of Florida
> Ph.D. Graduate Student
> 352-392-4032
> _______________________________________________
> AMBER mailing list
> AMBER_at_ambermd.org
> http://lists.ambermd.org/mailman/listinfo/amber
>

_______________________________________________
AMBER mailing list
AMBER_at_ambermd.org
http://lists.ambermd.org/mailman/listinfo/amber