Why it is needed?
Different MPI implementations have their specific benefits and advantages. So in the specific cluster environment the HPC application with the other MPI implementation can probably perform better.
Intel® MPI Library has the following benefits:
Intel MPI 4.1.3 on RHEL6.4: trying to bind ranks in two simple fashions:(a) 2 ranks to the same processor socket and (b) 2 ranks to different processor sockets.
Looking at the Intel MPI Reference Manual (3.2. Process Pinning pp.98+), we should be able to use options in mpiexec.hydra when the hostfile points to the same host
-genv I_MPI_PIN 1 -genv I_MPI_PIN_PROCESSOR_LIST all:bunch
-genv I_MPI_PIN 1 -genv I_MPI_PIN_PROCESSOR_LIST all:scatter
First i have to provide this information:
1- i have instaled the latest version of Intel MPI.
2- i have to use it through Ansys HFSS 15 x64 which is a EM-software.
3- HFSS dont have any problem with discrit processes (for ex. 15 paralel process will share trough 3 computer on network correctly)
4- i need to use memory of other computers on the network, so need distribute the RAM usage.
5- the error i got every time is "authentication faild" or "unable to create child process in hosts" (or somthing like these)
Dear developers of Intel-MPI,
I found, that the helpful option -prepend-rank does not work when launching a parallelized Ftn-code with mpirun when using INTEL MPI-5.0 :
mpirun -binding -prepend-rank -ordered-output -np 4 ./a.out
The option actually has no effect with INTEL MPI-5.0 (with INTEL MPI-4.1 it worked). No rank-numbers are prepended on the display to the output lines of the program.
Dear developers of Intel-MPI,
First of all: Congratulations, that INTEL-MPI now supports also MPI-3 !
However, I found a bug in INTEL-MPI-5.0 when running the MPI-3 shared memory feature (calling MPI_WIN_ALLOCATE_SHARED, MPI_WIN_SHARED_QUERY) on a Linux Cluster (NEC Nehalem) by a Fortran95 CFD-code.
I have a cluster with two kinds of nodes joined into parallel calculation: the first kind is the nodes with 2 CPUs and 4 cores in every CPU, the memory in every node is 32 GB, the second kind is the nodes with 4 CPUs and 8 cores in every CPU, the memory in every node is 256 GB. All nodes have Windows Server 2008 HPC in stalled and they are all joined into one domain controlled by another node (which is not joined into the calculation). I launched the job by the following command:
I noticed that Intel has release Intel MPI 5.0 and I downloaded latest version to test it. The command I used is as follows (it works with latest Intel MPI 4.1.x):
mpiexec -wdir D:\Users\tang\Debug -mapall -hosts 1 n01 1 D:\Users\tang\Debug\fem
However, the following error messages displayed:
Intel® Cluster Checker version 2.2 is an update in conjunction with Intel® Cluster Ready Architecture Specification version 1.3.1 release and also adds additional functions and capabilities.
Version 2.2 includes:
I am testing Intel MPI 4.1 with test.c (the provided test program).
Whenever I run > 2000 ranks the program executes correctly but fails to end gracefully.
mpiexec.hydra -n 2001 -genv I_MPI_FABRICS shm:ofa -f hostfile ./testc
It stalls at
Hello World: Rank 2000 running on host xxxx
##<stalls here; does not return to command prompt>
(If I use -n 2000 or less, it runs perfectly.)
I have testing 3000 ranks using OpenMPI, so it doesn't seem to be a cluster/network issue.