Self-hosted router - how to optimize Kubernetes and what is the limit?

We are using router/supergraph and we were wondering how many super graphs we need for a high load event (like a Super Bowl ad). How many concurrent connections per machine? How have 0ther people scaled. I think # of sockets and TIME_WAIT may be an issue. Unfortunately we are using EKS on Fargate and there does not appears to be a way to get networking sessions on Linux without privileged mode.?