The main reason PMTiles is one file and not two or more files is that it enables atomic updates in-place (which every mature storage platform supports) as well as ETag content change detection in downstream caches. All of the server and serverless implementations at http://github.com/protomaps support this now for AWS, S3-compatible storage, Google Cloud, and Azure.
Also, I'm pretty sure range requests are going to be difficult to cache. That implies going to origin every request which is bad.
First 100KB of a 100GB+ file:
curl -H "Range: bytes=0-100000" https://overturemaps-tiles-us-west-2-beta.s3.amazonaws.com/2... --output tmp -w "%{time_total}"
First 100KB at the 100GB mark:
curl -H "Range: bytes=100000000000-100000100000" https://overturemaps-tiles-us-west-2-beta.s3.amazonaws.com/2... --output tmp -w "%{time_total}"
BTW, we are discussing latency with bdon in this issue, it seems to be specific to Cloudflare: https://github.com/hyperknot/openfreemap/issues/16
> What about PMTiles?
> I would have loved to use PMTiles; they are a brilliant idea!
> Unfortunately, making range requests in 80 GB files just doesn't work in production. It is fine for files smaller than 500 MB, but it has terrible latency and caching issues for full planet datasets.
> If PMTiles implements splitting to <10 MB files, it can be a valid alternative to running servers.
[1] https://github.com/hyperknot/openfreemap