In Introducing FlowProxy, Chainbound collaborated with Flashbots to deprecate its original orderflow proxy implementation in Go in favor of a new one built in Rust, created with the goal of reducing end-to-end latency, improving efficiency, and increasing observability in BuilderNet’s networking layer. This first collaboration aimed at bringing it to production-ready quality and ready for deployment, already observing good improvements in networking and processing latency.
This second collaboration expanded on the Next Steps section outlined in the previous report, with success. It includes a quantitative analysis of order losses and latencies across BuilderNet, and extensive work to improve the transport layer to reduce both CPU and memory usage. We encountered a couple of non-obvious learnings that we outline below, and we hope this will be useful to the community.
Since the article contains interactive graphs, I’ll link to it here: FlowProxy: Approaching Optimality | Chainbound Blog