Several different users of AWS Parallel Cluster are doing some CFD case design and running smaller compute simulations on their desktop machines, using the CFD software GUI, and then designing and submitting more complicated simulations to Parallel Cluster using slurm as the complexity of the case exceeds certain complexities / mesh sizes greater than their systems can handle.
We also have a related requirement for running the software, and some other visualisation software via a GUI desktop interface on a system that has realtime access to the cluster's shared storage resources, for monitoring job progress and execution.
The NICE-DCV integration available for the head node is perfect for this, and it is what we are currently doing, however this has several drawbacks:
- We don't really want users accessing the head node and running application workloads that may compete with and interfere with running jobs
- We often temporarily need some much larger GPU instance types (eg. 128GB RAM) for this work, which gets expensive running the head node 24/7
- We want to avoid interruptions to the head node that are required to resize it back down to normal
Feels like the perfect use case for the Parallel Cluster Login Node feature but as far as I can tell they do not support NICE-DCV, only SSH access.
Am I missing something or is there a better way to achieve this?
AWS OFFICIALUpdated a year ago