It would be hard to give very specific examples, but here we scale the GB Cloud Proxy (which implements the GB Proxy codebase) based on both CPU and Memory utilization, both with a target value of 80%. Currently we are able to balance the load by autoscaling between 2-6 ECS instances. I am sure that we will need to tweak this as we roll out to more customers, but it's a good starting point I think.