It would be convenient to be able to do open_files on a list of HTTP URLs. This cannot be a complete file-system like the other backends, because there is no possibility of glob(), but I still think it would be useful for accessing remote data.

In general, the size of a given URL's data can be found through a HEAD command, and chunks read using the Range header keyword; some servers do not support that, however, and then downloads could only be parallelised across multiple files.