Performance issues can be for different reasons:

• Misuse of the queue manager. When the job is launched it must be specify the number of processes the job must use, if the number is incorrect it can affect job performance and the other users too.
• Improper implementation and processes execution on nodes. For example, it is possible that you are using message passing in a single node, when it would be better shared memory.
• Inappropriate storage use. If high-performance storage at /scratch is not used, you can experience an I/O performance decrease.
• Inappropriate network communication use. If processes are running on both nodes and communications are needed between them, ensure you have not specified the name of the nodes as cn001 and cn002. By default, it uses the computing network, so it is not necessary to indicate the nodes name to run in. If needed, it must be specify the name as cncp001 and cncp002.

After checking the above, if problems persist, please contact the CénitS technical team.