My (VG) personal opinion is that DHT is a pure attractor of academic interest lacking practical value. To elaborate a bit, DHT...
- is not that decentralized anyway (Khashmir employs bootstrap nodes, e.g. router.bittorrent.com:6881 - how is it better than DNS roots?)
- its main problem is the fact peers have to keep unchecked data created by some unknown peers and consumed by some other unknown peers (by "unknown" I mean untested as opposed to e.g. currently unchoked peers who are reachable and tested)
From the practical standpoint, that creates the discussed problems of stale entries, unreachable peers etc. Considering incomplete connectivity in the current networks, any measures of DHT maintenance seem expensive and unreliable.
- Don't be more saint than the Pope. Use the mainline code. The only improvement that seems worthy is the faster discovery of first k peers to bootstrap PEXing.
- Do not put heavy load on DHT; use it to find peers in external swarms.
- Test/study/improve PEX as the main decentralized tracking mechanism. (What about extended PEX on the overlay swarm? I mean, to track all the little pieces of content: previews, subtitles, etc.)