Hi,
My users find HDF5 really useful (thanks :), so I'm trying to build the parallel version against the various compilers and MPIs on our RHEL6 box. Unfortunately, I've hit a problem with one particular combination:
HDF5 1.8.9
MVAPICH2 1.8
PGI 13.6
HDF5's "make check" doesn't complete - it gets stuck on t_cache. "top" shows 6 t_cache processes consuming CPU and never complete. I don't seem to have this problem if I use a compiler from a different vendor (GNU or Intel), or the same compiler but with intelmpi or openmpi.
In case it was a fixed bug, I tried the latest releases of each component, with the same result:
HDF5 1.8.11
MVAPICH2 1.9 (derived from MPICH-3.0.3)
PGI 13.6
Build notes and output of t_cache.chklog is appended below.
Does anyone have any advice, please?
Thanks,
Mark
···
--
-----------------------------------------------------------------
Mark Dixon Email : m.c.dixon@leeds.ac.uk
HPC/Grid Systems Support Tel (int): 35429
Information Systems Services Tel (ext): +44(0)113 343 5429
University of Leeds, LS2 9JT, UK
-----------------------------------------------------------------
MVAPICH2 build notes:
./configure --prefix=$prefix --with-rdma=gen2 --enable-shared \
--enable-romio --with-file-system=lustre+ufs
make
make install
HDF5 build notes:
export CC=mpicc
export CXX=mpicxx
export FC=mpif90
export F77=mpif77
export F90=mpif90
./configure --prefix="$prefix" --enable-parallel --disable-shared --enable-fortran --with-zlib --with-pic
make
make check
t_cache.chklog:
============================
t_cache Test Log
===================================
Parallel metadata cache tests
mpi_size = 6
express_test = 1
*** Hint ***
You can use environment variable HDF5_PARAPREFIX to run parallel test files in a
different directory or to add file type prefix. E.g.,
HDF5_PARAPREFIX=pfs:/PFS/user/me
export HDF5_PARAPREFIX
*** End of Hint ***
5:setup_rand(): seed = 125651.
0:setup_rand(): seed = 125675.1:setup_rand(): seed = 3:setup_rand(): seed = 125672.4:setup_rand(): seed = 125678.
125673.
2:setup_rand(): seed = 125679.
Testing server smoke check PASSED
Testing smoke check #1 -- process 0 only md write strategy