Kino runs a Rack 3 app on parallel Ractors in one process, and reports about 4x lower memory than a Puma cluster on Rails at roughly a fifth of its throughput

Sources: Kino repository, HN discussion
Summary: Kino is a Rack 3 server that runs request handling on parallel Ractors inside one process instead of forking worker processes. The reported Rails comparison against a Puma cluster is roughly 4x lower memory at roughly a fifth of the throughput. The reason the throughput side is weak is stated in the project's own text: Rails is not Ractor-shareable, so it falls back to Kino's threaded mode on a single GVL-bound process. The figures are the project's own and no independent reproduction was located this run.
Why it matters: Rails is not Ractor-shareable, so it runs only in Kino's threaded fallback on one GVL-bound process, which is where the throughput gap against a fork cluster comes from.