A systems project for exploring the behavior of feed handlers and downstream consumers under uneven traffic.
What it does
- emits synthetic market-style events with configurable burst patterns
- measures throughput and queue depth under different consumer strategies
- compares parsing, batching, and dispatch approaches
- logs median and tail latency for each stage in the path
Why I built it
I wanted a controlled environment for testing low-latency assumptions without needing access to a production data source. This makes it easier to reason about backpressure, message sizing, and where latency variance actually comes from.
What I am learning from it
- small memory decisions can create visible downstream effects
- throughput alone is a weak metric without tail-latency visibility
- instrumentation has to be cheap enough to keep enabled during iteration