To use HTTP retrievals, you only need to point your config file to an HTTP endpoint that hosts the epoch-NNN.car file, which supports HTTP range requests.
If your remote storage solution cannot handle large monolithic files like the 600 GiB files we produce, you can also use split retrievals in HTTP mode, similar to how they work in Filecoin v1 retrievals.
You can use go-fil-dataprep to split the files in any size you want. This tool will also generate metadata.yaml. After you have generated metadata.yaml and the split, you need to create a map from this tool's “piece CIDs” and the URI of the specific pieces.
A sample config file for this use case would look like this (NOTE: this is just a sample; we currently do not host split retrievals on https://files.old-faithful.net):