Semi-distributed mirroring

[mdlbear]

This came up in comments to a post by ravan explaining why she's not a member of MeetUp.com anymore, and wondering whether it would be possible to set something up as a federation of connected volunteer sites. It would want to be a federation mainly to save bandwidth and storage costs. Ravan asked, "Unfortunately, I'm not familiar enough with P2P stuff to code it, but... mdlbear? Is this doable easily?"

here are my thoughts on the subject, slightly edited.
It wouldn't really be P2P; more like local mirroring except that, if you did it by region, you wouldn't even have to mirror very much. Sort of like what CPAN and CTAN do, or SourceForge's download mirrors.

Basically you'd have a single main site, www.example.com, that everybody goes to first. That would let them pick a region of interest and redirect them to, say, sfba.example.com or nyc.example.com, both of which would be CNAME aliases for whoever was volunteering bandwidth in that area. Big population centers like the NYC or SF Bay Area would probably have to be further subdivided, perhaps by interest.

Login cookies would be valid across the entire domain, so users would only have to register in one place; the registration database and user profiles are the only things you'd have to replicate across the entire domain. (Actually, if every user has a home region, you'd only have to replicate names, passwords, and regions. Piece of cake.)

You'd do "backups" by replicating each region's information on only a small number of "buddy" sites, probably using rsync. I like to use flat files for everything, with a short file for every entry, because directory updates are atomic. You'd then use something like blosxom or nanoblogger to (statically) roll up RSS feeds, "what's new" lists, and so on.

Can you tell I've thought about this kind of thing? In particular I've been thinking about using it for PenguinSong.net, which I haven't done anything with recently but have been thinking of turning into a federated network of songwriters' sites (since the original idea, a centralized co-op, would be too much effort and probably wouldn't work out). This kind of thing would work for any kind of federated network of sites -- the main thing is to divvy the content up so that no one site has to have the whole thing.

Comments

[eleccham.livejournal.com]

I don't think you'd want to break big cities down by interest; some interests will generate a lot more bandwidth (popularity) than others, and it'd quite possible be hard to predict ahead of time.

Perhaps the "backup" sites could be load-balanced?

[mdlbear]

Presumably one would do it adaptively -- start out all in one place, then break it down later as the load increased.