Dynamic segmentation
The file is never split in advance.
Naïve accelerators cut a file into equal pieces up front. When one piece is served
slowly, the whole download waits for it. This engine never does that.
It starts as a single segment. Each new connection finds the largest remaining
piece and splits it in half, taking the back end for itself. When a connection finishes, it
adopts an abandoned segment or halves the biggest one again. The fast connections keep
taking work off the slow ones, right up until the last byte.
- No fixed segment count — the shape adapts to real conditions
- Slow connections get helped, not waited on
- A split floor stops pointless division of tiny pieces
- Direct positioned writes — segments land at their real offsets, so there is no reassembly pass at the end