Blog — Intrane.fr — #performance

Product · August 2026

How I Made remotecmd 1.7× Faster Than scp Over a 77ms Relay

remotecmd routes file transfers through a WebSocket relay — which means a single TCP stream can't fill the pipe on high-RTT links. I spent two days reading kernel source, CERN GridFTP papers, and gorilla/websocket internals. The result: parallel TCP streams that auto-tune by file size, atomic header+binary framing in the relay, and gzip writer pooling. 50MB over 77ms RTT: 1.74× faster than single stream, 1.21× faster than scp. Here's what worked, what didn't, and why.

#remotecmd #performance #tcp #websocket #parallel-streams #open-source #devtools

The Port That Found a 30,000× Bug in My Own Language

SuperCLI already ran in Node.js and Zig. I ported it to my own language, machin, to dogfood — 10,158 plugins, one 69 KB binary, 24 tests. Then the live test hung for 30 seconds. The bug wasn't in my code; it was in my language's string model. Fixed it: 30,000× faster. This is why you dogfood.

#machin #supercli #dogfooding #performance #ai-engineering

A 1B-parameter LLM at 20 tokens/second, in a language you've never heard of

I challenged machin — my machine-first language — to run TinyLlama-1.1B at 20 tok/s on a laptop CPU, pure MFL, no C kernels. It hit 21.8, token-identical to llama2.c, and 4x faster than the reference C single-threaded. Here's the 2.1 -> 21.8 arc and what it drove into the compiler.

#machin #llm #inference #performance #ai-engineering