Wasn't so much users are traffic. We had Los Angeles, Montreal, Helsinki, Sydney, Buenos Aires, and Montreal data centers with GeoDNS routing. Each relay synced with the others to have consistent data. While also syncing with about 10 of the other top relays to replicate their data. The current LMDB for strfry is 152GB (will be backing up and pruning soon). Was planning to have a separate "archive" relay for historical data (now on-hold).  

 feels like very very premature scalability  

 Was always a PoC and consist performance example. Most relays are single-point failure be it one server or all on Cloudflare (for example). One year ago was handling a lot of all nostr traffic one way or another. 

 but isn't the idea to have many relays ran by different people? instead of one giant immense super scalable relay