I have some doubts regarding the Intel software studio for parallel arch and the Brazilian seller is not able to answer. I need to solve these doubts before buy the Studio for my company. Can somebody help me?
1- Currently we are using OpenMPI. Which advantages Intel MPI provides over OpenMPI?
2- OpenMPI error handling is not good. The MPI Lib from Intel is better for error handling and recovering? For example, if one rank in my mpi comm world dies how can I handle this using Intel lib?
3- Currently we use GCC. Intel compiler is better? We are running in a cluster with several nodes, with MPI doing the communication between the nodes.
Any other recommendations? We host our application at Amazon.
Thank you all in advance!
These questions aren't simply answered.
You don't even mention having perused Intel white papers which bear on the questions you ask. Generally, they compare recent releases of Intel software tools against older versions of OpenMPI and gcc supported by "stable" linux distros such as Red Hat.
If you want to take full advantage of current OpenMPI and gcc, you must take on more maintenance and update tasks yourself, some of which are done for you in exchange for Intel licensing fees. For example, most medium to large installations would install a recent gcc alongside the one which comes with the linux distro, and build their own copies of a current OpenMPI. Larger installations would support several MPI implementations, including OpenMPI and Intel, typically using the modules facility to select among supported alternates.
Intel MPI has supported automatic process pinning for several years; more recently, OpenMPI makes a similar thing available.
MPI isn't particularly friendly for error recovery; I wouldn't say there is a great difference among implementations, but more specific questions about Intel MPI would be topical on the HPC and Cluster forum, which gets more attention from Intel MPI support people than this forum does.
Thanks for your answer.
Yep, MPI is sometime hard to handle, but is a great tool and is perfect for our application. Looks like Intel MPI is faster than OpenMPI and more madure. I'll check these white papers.
Any other reference I should take a look before buy?