50k RPS on 1 GB RAM

Write-up in progress. This is a stub so the URL exists.

I built a telemetry stack that sits in front of production traffic and holds ~50,000 requests per second on a machine with 1 GB of RAM.

On paper that is the wrong shape: heap, GC, buffers, TLS, cardinality — pick one and you are already over budget. In production it holds. The trick is not a bigger instance.

What the full note will cover:

• Where the bytes actually go (and where they must not)
• Why a 1 GB box is the constraint, not a flex
• Hot path: no alloc if we can help it, ring buffers, drop policy that doesn’t lie
• HTTP/3 / QUIC on the edge when it is cheaper than another TLS 1.2 worker
• What breaks first when you try this the obvious way

If you need the numbers before the essay ships, mail me.