I have a professor from one university asking the following question:
"Getting VTune installed and running on the cluster will apparently require significant modifications including installing drivers on all nodes. We are trying to weigh cost/benefit of installing these tools and were hoping that you or someone you could point us to could help us find the easiest and least invasive way to install them"
Could any Vtune expert in this forum give some info if Vtune can be used in a computer cluster environment?
Well, don't call me an expert; I haven't had much success with multiple node VTune runs.
Yes, the VTune driver has to be installed on each node. As the PTU (download from whatif forum) provides this functionality and allows you to generate a command line from the GUI, that command line can be run as a cluster distributed job, collecting data for each node separately.
For many applications, it's sufficient to run MPI under VTune on a single node. If the application doesn't distribute the work evenly enough across nodes, VTune isn't necessarily the way to diagnose it.