Long lived channel scailability

[vallej...@gmail.com]

I have a scenario where I need to keep channels open for a long time to tens possibly hundreds of thousands of endpoints. I'm evaluating gRPC for this use. I know I will need a horizontal scailability in the system but I'm wondering how many long lived channels a single machine can manage. I know this can depend on various factors including but not limited to machine specs, workload characteristics etc, but I'm curious if any research / testing has been done on similar problems before. Does anyone have ballpark estimates on what the limits might be?