Cloudflare and Akamai proposed optional dictionary-compression for HTTP [0] that could help smaller files achieve better compression ratios than gzip/deflate [1].

Given the fact that zipf distribution holds true for natural languages [2], dictionaries can be a simpler solution to text compression in low-power / compute environments?

[0] https://tools.ietf.org/id/draft-vkrasnov-h2-compression-dict...

[1] https://blog.cloudflare.com/improving-compression-with-prese...

[2] https://www.youtube-nocookie.com/embed/fCn8zs912OE

I'm able to follow the article right up to the point where the author talks about replacing streaming with dictionary compression. How does this work, exactly? Don't you still need the same number of back and forth messages, since the protocol doesn't change?

Yes, the number of messages is the same. For small messages, per-message overhead may swamp the benefit of compression.

Discussed at the time: https://news.ycombinator.com/item?id=16398752