Performance consideration for search cluster

Hello,
We have a large catalogue of about 120,000,000 documents and we want to release these for general public consumption from different devices.

What kind of things would I need to consider when designing the cluster purely for search at these volumes?

Would reverse proxy have any performance implications, if this is added?

Any advise is much appreciated

Depending on the load, I'd may be increase the number of replicas and may be the number of nodes to be able to deal with the search requests.

Having a coordinator node in front of the cluster would also help may be.

BTW did you look at cloud.elastic.co for this?