[gmx-users] Water Bug

[Hank deHaan]

Hello

I have a system consisting of 4064 Protein atoms
                               845 Water molecules
		               and some Ions

I've compiled Gromacs 3.1.4 on a Beowulf of Alpha/Dec processors 
running Linux.  While the system runs well on a sinlge node, 
when I try to run across even two processors I get the following 
error message:

prun: /home/hdh/G3_hdh/ev6/bin/mdrun_hdh (pid 2705) killed by signal 8 
(SIGFPE)
prun: generating backtrace for /home/hdh/G3_hdh/ev6/bin/mdrun_hdh 
core.2705
#0  0x120066b74 in angles ()
prun: dumping elan exception state for /home/hdh/G3_hdh/ev6/bin/mdrun_hdh 
core.2705
edb: found elan3_exceptions struct at 0x200002c9558
edb: exceptions from '/home/hdh/G3_hdh/ev6/bin/mdrun_hdh' corefile 
'core.2705'

I thought this might be the bug involving having optimized water on node 
zero, but I seem to run into simliar problems whether the system is fully 
hydrated, dehydrated, or selectively hydrated (similiar in the sense that 
sometimes the exception is in the function do_14 or print_ebin)

Also - the system has cylindrical symmetry - so dividing the job over just 
two nodes should ensure that there is some water on node zero (providing 
that Gromacs partitions the system by geometry - as  I think it does) - 
and yet it still crashes

Speaking of the water-bug, I've read up in the mailing list archives a bit 
and was wondering if it will be addressed in the next release of Gromacs.


Thanks for any advice you can provide,

                                         Hendrick de Haan