- Mark as New
- Bookmark
- Subscribe
- Mute
- Subscribe to RSS Feed
- Permalink
- Report Inappropriate Content
My customer asks:
For use on a Linux cluster, is it sufficient
to get the one-seat license? We can put it on only one of the nodes, so users
have to login to that node for the compilation of their program, and launch the
cluster run from there. How many processes can be used for the (mpi) run - is
there a limit? Can a run be for the hybrid application, with mpi between the
nodes and shared memory programming on each of the participating nodes? Is
there a limit on the number of threads generated on the nodes?
Please advise,
Thanks.....Dave A. from Lifeboat Distribution
Link Copied
3 Replies
- Mark as New
- Bookmark
- Subscribe
- Mute
- Subscribe to RSS Feed
- Permalink
- Report Inappropriate Content
You'd need to study the implications of the type of license (single user, floating, node locked, ....).
It looks like the license limit you should be concerned about violating is in the use of one or more individually licensed Intel compilers by multiple people. You could take care of that by adding a floating license for the relevant compiler. Then you wouldn't be limited by considerations of by whom and where the build is done. If it's a floating license for Cluster Studio, you should have no problem.
Intel MPI is well suited for MPI hybrid operation with Intel OpenMP; it has automatic default placement of threads local to CPUs according to OMP_NUM_THREADS. This works only when each node is devoted to a single job; you would have to turn off thread pinning if the nodes are shared without coordination, but you can't expect consistent performance in that case, particularly if there are more threads than hardware contexts. You would run out of hardware resources before you hit a fixed limit on number of threads.
Open source MPI implementations tend to have defaults which are compatible with running more MPI processes than hardware resources; for Intel MPI you must study how to turn off pinning for such cases. This is one of the reasons why job schedulers are normally used to assign nodes to 1 job at a time (torque/Maui et al.).
If you look up Top500 results you will see demonstrations of rather large number of MPI processes, number of threads per node limited to the number of physical cores (until a MIC result may appear).
It looks like the license limit you should be concerned about violating is in the use of one or more individually licensed Intel compilers by multiple people. You could take care of that by adding a floating license for the relevant compiler
Intel MPI is well suited for MPI hybrid operation with Intel OpenMP; it has automatic default placement of threads local to CPUs according to OMP_NUM_THREADS. This works only when each node is devoted to a single job; you would have to turn off thread pinning if the nodes are shared without coordination, but you can't expect consistent performance in that case, particularly if there are more threads than hardware contexts. You would run out of hardware resources before you hit a fixed limit on number of threads.
Open source MPI implementations tend to have defaults which are compatible with running more MPI processes than hardware resources; for Intel MPI you must study how to turn off pinning for such cases. This is one of the reasons why job schedulers are normally used to assign nodes to 1 job at a time (torque/Maui et al.).
If you look up Top500 results you will see demonstrations of rather large number of MPI processes, number of threads per node limited to the number of physical cores (until a MIC result may appear).
- Mark as New
- Bookmark
- Subscribe
- Mute
- Subscribe to RSS Feed
- Permalink
- Report Inappropriate Content
Hi Dave,
Basically, what Tim said. The only restrictions you should encounter are those in the EULA. As for the Intel MPI Library and Intel Trace Analyzer and Collector, the runtime components are not limited, the development and analysis components are. If you are using a single user license, there should be no problem. If this is a floating license, then only one person can use the development tools at a time. Please have the end user consult with their legal team or ours if further clarification is needed.
Sincerely,
James Tullos
Technical Consulting Engineer
Intel Cluster Tools
Basically, what Tim said. The only restrictions you should encounter are those in the EULA. As for the Intel MPI Library and Intel Trace Analyzer and Collector, the runtime components are not limited, the development and analysis components are. If you are using a single user license, there should be no problem. If this is a floating license, then only one person can use the development tools at a time. Please have the end user consult with their legal team or ours if further clarification is needed.
Sincerely,
James Tullos
Technical Consulting Engineer
Intel Cluster Tools
- Mark as New
- Bookmark
- Subscribe
- Mute
- Subscribe to RSS Feed
- Permalink
- Report Inappropriate Content
Thank you both, I will relay to the customer.
Dave A. from Lifeboat Distribution
Reply
Topic Options
- Subscribe to RSS Feed
- Mark Topic as New
- Mark Topic as Read
- Float this Topic for Current User
- Bookmark
- Subscribe
- Printer Friendly Page