Bursting to Cloud from On-Prem HPC Cluster

When is it appropriate to burst to cloud from an on-prem HPC cluster? Are there certain metrics and ranges for those metrics that might serve as triggers? What factors go into a decision to ‘burst to cloud’?