Multiple subdomains drastically slow down simulation

[Xavier Quinn]

I am trying to run a simulation over an ssh cluster on a pair of Jetson TX1s, but when the simulation has multiple subdomains the MLUPs drop to one fifth of what they where combined across the subdomains. This happens when the simulation is run on the cluster as well as on one machine, so we know it is not an issue with communication speed.

Any assistance would be appreciated, thank you.