I would like to thank you first for reading my Thread. I will try to be as precise as possible.
I am an Intel Visual Fortran 10.0.0.27 for Windows(happy) customer. Our programs are mainly a mix between Fortran (Compiler Intel)and C++ (Compiler VS 8.0 from Microsoft). We only use a single processor machine to run programs.
In a context of a future parralelization of our simulation softwares on distributed systems,I reached the " cluster OpenMP for Intel Compilers" page which leads me to the following questions:
- Existing code will be mainly splitted in 2 parts, these two parts communicate mainly via COMMON variables. We search for a system which could helpthe 2 parts tocommunicate using the COMMON variables (shared memory)in a distributed architecture. If we have well understood the aim of "Cluster OpenMP", that's exactly what we need. Am i right? (Maybe description is too light, i can develop if needed.)
- As Windows customer, i haven't seen any windows version of "Cluster OpenMP", is it available? Is it included in Intel
Cluster Toolkit 3.1 (windows)?
- As explained in point 1., we use a mix between C++ (VS 8.0 microsoft compiler) and IFORT 10. Is "Cluster OpenMP" working with this duet, do we need to use a C++ Intel compiler to work?
The Intel Cluster Toolkit does not include Cluster OpenMP for IntelCompilers. I don't think thatCluster OpenMP is available on Windows platform. Please get more information at http://software.intel.com/en-us/articles/cluster-openmp-for-intel-compilers
I am interested in the Cluster OpenMP for Windows too. Is there any good news about this product?
Now I have some problems to drive many external programs (Ansoft Maxwell) on distributed machines with multiplecore on every machine. I have thousands of projects to be simulated by Ansoft Maxwell so I wish all machines work at the same time with many Ansoft Maxwell launched on every machines.
Now I have two programs to drive the Ansoft Maxwell. One is MPI based and the other is OpenMP based. The OpenMP based program can let many Ansoft Maxwell programs launch and work at the same time correctly, but the MPI based program can't. The problem is that after the MPI based program launched many Ansoft Maxwell programs for about 1 second, only one program can keep running and all other programs are killed.
Since I found that the Intel have the Cluster OpenMP product, I wish it can help me to solve this problem. But now there is only the Linux version, when could the Windows version comes?
Among the more recently introduced parallelism models which may deal with some of the same requirements is the co-array Fortran which was introduced in recent Intel compilers.
If you are running independent jobs on a network, it doesn't look like any of the methods you mentioned is applicable. Perhaps a cluster queuing system such as PBS or a network application such as LSF or Condor.