Re: Names: Decentralized, Secure, Human-Meaningful: Choose Two Thomas Lord 27 Jan 2006 17:28 UTC
(Sorry about the top-posting. I use an off-the-shelf collection of GNU/Linux software from a major vendor and somewhere in the stack there is a long-standing bug that prevents replies to lists from working correctly in my (rather banal) environment.) On Fri, 2006-01-27 at 12:45 +0000, Tony Garnock-Jones wrote: > Your post, and Tom Lord's note on the politics of naming, reminded me > of an excellent essay introducing "Zooko's Triangle" [1], laying out > the tradeoffs between decentralization, strong identity, and > human-meaningful naming. (Another short essay vaguely on topic is [4].) > Nice presentations. Thank you. [1] http://www.zooko.com/distnames.html > [2] http://www.skyhunter.com/marcs/petnames/IntroPetNames.html > [3] http://ciphergoth.livejournal.com/110893.html > [4] http://shirky.com/writings/domain_names.html > [5] http://www.cc.gatech.edu/~tony/L/ > [6] http://www.cc.gatech.edu/~tony/L/#Protocols%20and%20Default% 20Behavior > I really like the rhetorical device of the triangle (first two links). I encounter design tensions like that all the time -- it's nice to have more symbolic devices for communicating them. Someone else also sent me this link (recommended): Ludovic Courtès: > The proposal at the end of your reply reminds me of SPKI/SDSI's > name spaces, described at the beginning of: > Clarke et al, ``Certificate Chain Discovery in SPKI/SDSI'', 2001, > http://theory.lcs.mit.edu/~rivest/ClarkeElElFrMoRi- CertificateChainDiscoveryInSPKISDSI.ps > The idea, roughly, is that each one can have identifiers in their > local name that may be bound either to a name in another person's > name space, or to some value (actually, a public key). Such a > binding is called a ``name'', precisely. Everyone can issue ``name > certificates'' for their local bindings, that is, cryptographically- > signed descriptions of the bindings. I became interested in names through work in revision control in the free software world. We wish to shun central authorities and other institutional environments when managing the corpus of free code but we also wish to have robust communities of discourse about and exchange of that code -- names are critical. I'm increasingly convinced that domain names should and *probably* will wind up being not much more important than phone numbers and only a little more important than IP addresses. Instead, we're going (knock on wood) to wind up with a "meta-net" -- a thoroughly virtual Internet to which new (virtual) nodes can be added without authorization from anyone and which is physically realized as a P2P layer over the real Internet. I don't mean the Really Big issues to derail the SRFI process but do think they are worth bringing to the conversation. -t