I am wanting to explore some other distributions to possibly use at work in the near future. One of the possible requirements will be to do an autonomous network install. Basically, that means the internet will not be reachable, and I would need to run any and all services within the isolated LAN. The obvious to get a network boot going would be the usual PXE components, DHCP, TFTP, and some network boot loader. Then a repository for the distribution would need to be hosted (like being a mirror). What I would like to see in a distribution is some obvious means to select just the files needed for one exact version and architecture (usually the latest version and for x86-64/amd64), load those on the netinstall server, and keep them in sync with rsync or similar tool. One server would be online to get this data, and a hard drive replica would be carried to the server on the isolated LAN (hard physical isolation which cannot get to the internet, even by proxy or VPN).

Basically, I'm wondering which distributions can handle that better. I don't want to mirror the entire distribution, though.

I'm not really sure what kind of response you're after here. It's not going to be about the distribution but the frame work to delivery them. I'd suggest looking at cobbler which can mirror all redhat and debian derivatives very well. As for the bit about not mirroring everything, that's a very subtle thing really, hiw do you know what you will and won't need etc.?

I'm not sure what I am looking for, either. I can imagine a few possibilities. Basically, something that can tell me WHAT I need to mirror for a specific version of the distribution (for example, Ubuntu 10.04 LTS, or Centos 5.5), and what portion needs to be mirrored to become a local repository (that I can add to the repository list and have it go there instead). The question "exactly what files do I need to make a complete working repository for this specific version" may be a better question to ask.

What I want to avoid is mirroring ALL the versions of Ubuntu, if all I want is (for example) 10.04. Likewise for other distros.

Note that I am not looking to make a public repository; we don't have the bandwidth to do that. But I am looking for something more than just a web cache (which would help on bandwidth issues). The reason for that is security containment (e.g. doing the installs in a LAN that will never have internet access, not even via a proxy). So, basically, I need to carry all the data over on a hard drive and when it's there, it provides all I need to do full installs, add many packages, and be as up to date as when I do the last mirroring to that drive. Mirroring by rsync or similar would be preferred to reduce traffic when updating.

For Ubuntu, this page seems to describe all the processes involved:

https://help.ubuntu.com/community/Rsyncmirror

But it also points out the problem of mirroring everything ... over 200 GB by now I am sure. One solution to what I am looking for is a description of what SUBSET of these files I will need for a specific version. That page refers to the location rsync://archive.ubuntu.com/ubuntu which if listed shows 4 subdirectories to choose from ... dists, indices, pool, project. The big question is how to avoid downloading what I do not need, while being sure I do get everything I do need, for the chosen version. It does not appear to be structured as "10.04" or "lucid" as a top directory to just do one sync to get that.