Infrastructure

Week 13 — Dublin instance

06 September 2026AmsterdamJordi Lluis

I ran the live bot on an AWS Lightsail box in eu-west-1 (Dublin), Ubuntu. I paid for it, logged into it, and later used it for daily recordings. I did not run the same clock in a second city, so this is not a region bake-off.

On the submit path I removed waits I could name: a warmed HTTP/2 client, coincurve signatures, token metadata cached off the hot path. Book updates could trigger an evaluation instead of waiting for the next poll. The per-market gate in code went from 50 ms to 5 ms. That last number is a throttle, not measured end-to-end latency.

The timer I still have starts at metadata preparation and stops when the API answers — prepare plus POST. On 3 May, StratB’s median on that clock was 35.35 ms. Most of those replies were resting orders. An acknowledgement is not a fill.

I deleted the VPS recordings. The public repo keeps the ledger script and the timing parser, not those days.

Standing up Lightsail shows I can run a process on a VPS and define a clock. It does not show that Dublin made the strategy better, and it does not explain the later loss.

The 3 May sample · Case study · Walkthrough

← Blog