Understand differences in how parallel file systems are set up and the impact on the performance results

We do not currently have details of how different parallel file systems are connected up to the HPC systems (e.g. how does the fabric connect into the interconnect?) and need to understand the differences here. Can we understand how these differences impact performance?