As I worked on ToCry I got curious about how well it performed.
It was NOT built with performance as the main goal. In fact it’s sort of guaranteed not to be the fastest way to do things. It uses sepia as data storage, which is the opposite of efficient, and I implemented it for a single user.
On the other hand, the best thing was to just write some scripts and measure things!
Turns out … it’s pretty fast? It can have about 20 simultaneous users and 300 RPS on a Pi 4 with 4GB of RAM, and industry standard says that equates to about 200 users.
200 users? On an old SBC!
Details here
FOOTNOTE: This is not scientific. What a “realistic” load is needs to be discovered by having actual users :-)