MPI_THREAD_FUNNELED,
MPI_THREAD_SERIALIZED, or
MPI_THREAD_MULTIPLE.
The previous versions of the Intel® MPI provide only MPI_THREAD_NONE and MPI_THREAD_FUNNELED levels in terms of the MPI-2 standard.
Is the Intel MPI Library fault-tolerant?
No. The only fault tolerant feature implemented at the moment is the ability to fall back upon the default sock device in case the Intel MPI Library device selected by the user at runtime cannot be initialized for some reason.
Set the I_MPI_FALLBACK_DEVICE environment variable to disable to prevent fallback. In this case if an attempt to initialize the specified fabric fails, the library terminates the MPI job.
Note: This is true for RDSSM or RDMA modes only.
Does the Intel MPI Library support parallel I/O calls?
Yes. The parallel file I/O part of the MPI-2 standard is fully implemented by the Intel MPI Library. File systems currently supported include the Unix File System (UFS) and the Network File System (NFS).
Does the Intel MPI Library support one-sided communication?
Yes. The Intel MPI Library supports both active target and passive target one-sided communication. The only exception is the passive target one-sided communication in case the target process does not call any MPI functions.
Does the Intel MPI Library support heterogeneous clusters?
No. The Intel MPI Library does not support clusters with nodes of different architecture or operating environment.
Does the Intel MPI Library support 32-bit applications on 64-bit operating systems?
Yes. The Intel MPI Library supports 32-bit apps on 64-bit operating systems on Intel® 64 platforms.
Can the Intel MPI Library handle a mixed 32/64-bit job?
No. The Intel MPI Library does not support heterogeneous configurations. The entire job must be IA32, Intel® 64, or IPF based.
Does the Intel MPI Library support the same user running MPD daemons simultaneously?
Yes. The Intel MPI Library supports it. Use the mpirun job launcher. It will create an independent MPD ring, start the requested MPI job over it, and clean up the ring when the job termintes. You can launch several MPI jobs simultaneously. It is also possible to use the MPD_CON_EXT environment variable. Set this variable to an unique values before each MPD ring starts.
Example 1:
login on cluster
mpirun -np numproc executable [args]
login on cluster
mpirun -np numproc executable [args]
Example 2:
login on cluster
export MPD_CON_EXT=ext1
mpdboot -n numnodes
login on cluster
export MPD_CON_EXT=ext2
mpdboot -n numnodes
Is there a Microsoft* Windows* version of Intel MPI Library?
Yes. There is Intel MPI Library 3.1 for Windows. Go to http://www.intel.com/go/mpi for more details.
How do I check a process pinning table?
Set the I_MPI_DEBUG environment variable to four. The Intel MPI Library will report process pinning information.
For example:
mpiexec -np numproc -env I_MPI_DEBUG 4 $executable
How do I verify the Intel MPI Library variables in effect?
Set the I_MPI_DEBUG environment variable to five. The Intel MPI Library will report effective Intel MPI Library environment variables.
For example:
mpiexec -np numproc -env I_MPI_DEBUG 5 $executable.
Operating System:
| SUSE* Linux Enterprise Server 9.0, Red Hat* Enterprise Linux 4.0, SUSE* Linux Enterprise Server 10, Red Hat* Enterprise Linux 3.0, SUSE* Linux* 9.x |
This applies to:
