In the "Intel MPI Library for Linux" installation guide (found here), it discusses installing the libraries on one node or multiple nodes of a cluster. How do I know which method to choose? What are the advantages / disadvantages of either kind of installation?
- Cluster Computing
- General Support
- Intel® Cluster Ready
- Message Passing Interface (MPI)
- Parallel Computing
As mentioned in the Installation guide IMPI supports two types of Cluster setup:
1) Install cluster node software on an automounted shared file system
2) Install cluster node software on every node of your cluster
The benefit of using a shared file system is that the installed components will be in a shared path and when you have to add/remove a node you can just mount/unmount the node on the head node.
For the second method you need to make sure that on each node, the Intel developer software components are sitting in the same absolute path when adding a node.
Thanks for your reply. So I'm assuming that if I chose the single-node installation, and it runs without errors on multiple nodes, there would be no benefit to performing the multi-node installation?
We suggest installing in a shared file system since its better both performance-wise and storage wise.
Having multiple having copies of installation consume more space compared to a single copy.
There are no major benefits of multi-node installation compared to the shared file system installation.
Have you chosen a way of installation?
Or do you need any more information?
Could you confirm If your original query is answered, so we can close this thread?