Zip Files You Can Jump Into Cost Only 1.6% Extra Space

Imagine a massive zip file — a photo album, a video, a pile of data. It takes up very little room, but to grab one photo you usually have to unpack the entire thing first. That is the trade-off compression has always forced on us: small files, slow access. This paper puts a number on that trade-off for the first time across four common formats.

The researcher took a 254 megabyte archive and cut it into small chunks you could jump to directly. The best format charged just 1.632% extra space for the privilege. The widely used "seekable zstd" format charged 6.57%. When the chunks got smaller — 4 kilobytes instead of 16 — the cost rose to 5.33% and 10.06%. In other words, paying a tiny bit of storage buys you instant access anywhere in the file. For comparison, 1.6% of a 254 MB file is about 4 MB — smaller than one phone photo.

That small cost unlocks handy things. Splitting a big archive into pieces becomes free, and sometimes actually saves space (-0.28% on one type of data). You can add new data to a file without reformatting it. Jumping to the end of a file takes exactly as long as jumping to the beginning. And the same file can be read by both a regular processor and a graphics chip — the hardware that powers AI — which matters for machine learning workloads.

The catch: this is a research paper, not a product launch. The formats exist in the lab, and the software you use daily (zip, gzip, standard zstd) will not change overnight. The author also lists 17 rejected ideas, including one that improved compression by 26% but was declined for other reasons. Expect this to trickle into cloud storage and AI tools over the next few years, not next week.