Zooko's Triangle

https://en.wikipedia.org/wiki/Zooko%27s_triangle

by mattwilsonn888 • 4 years ago

79 40 4 years ago

en.wikipedia.org

itake• 4 years ago

What about Facebook's onion domain?

https://facebookcorewwwi.onion/

This seems to be human meaningful, secure, and decentralized.

https://m.facebook.com/nt/screen/?params=%7B%22note_id%22%3A...

scraptor• 4 years ago

It's not actually secure and human meaningful. There are many possible domains at least as legit looking which are not owned by facebook, and so you must either have the machine remember that facebook is facebookcorewwwi.onion and not wwwfacebookcomyz.onion (secure) or trust the first domain that looks legit (human-meaningful).

chejazi• 4 years ago

I can't believe they managed to grind that one.

arwt2• 4 years ago

I'd love to see some insight as to how they were able to generate that domain name. Surely if they can do that (years ago, even), then a well-funded adversary would be able to generate the same now, given enough time.

Maybe it took x thousands of compute years to generate the secret key for `facebokcorewww` - and they didn't care about the last character. But still, let's say you're a government agency with endless resources - how hard would it be to recreate that private key? If a private corp can do it once with finite resources - why can't you?

Was it is just a stroke of luck for those working on it? What are the chances?

pseudalopex• 4 years ago

They stress tested a compute farm for by brute forcing onion addresses. About 10 per week per CPU started with facebook. corewwwi was their favorite.[1] Onion addresses are longer now.

[1] https://alecmuffett.com/article/15996

cowtools2• 4 years ago

It is not secure by merit of the fact that they were able to brute-force it. The v2 onion domains are weak. the v3 onion domains are much longer.

haakon• 4 years ago

Also worth noting that v2 domains are phased out now and will no longer work. Facebook's new onion address is facebookwkhpilnemxj7asaniu7vnjjbiltxjqhye3mhbshg7kx5tfyd.onion, which is not exactly human-meaningful.

mwilcox• 4 years ago

21e8 fixes this

ronreiter• 4 years ago

Blockchain is an obvious counterexample

wikfwikf• 4 years ago

But aren't blockchain solutions a form of centralization?

The blockchain itself has a decentralized structure. But when we use it as a basis for our naming scheme it becomes a single point of failure. It isn't possible for some participant in the naming scheme to choose to use a different blockchain which they prefer.

cjbprime2• 4 years ago

Zooko's Triangle was a useful lens, but I think of it has having long been solved (or, refuted) by many blockchain projects.

amingilani• 4 years ago

I concur, for example, ENS domains are memorable, secure and the whole contract is decentralized. And you can send Vitalik eth at vitalik.eth

googlryas3• 4 years ago

Woe betide other Vitaliks!

mindslight2• 4 years ago

IMO, blockchain human-readable naming generally creates a single coherent namespace and are thus "centralized" in terms of Zooko's Triangle. The singular authority is vested in shared consensus, but is singular authority nonetheless.

The real answer to Zooko's Triangle is to do human-meaningful name lookups as little as possible. For instance, in an ideal world I should be able to run a nameserver that stuffs all of ICANN's TLD pollution under an .icann label, such that I could type "blog.google.icann" to access what is currently called "blog.google". References to other resources would be specified with decentralized non-human-readable (ie cryptographic) identifiers, and outgoing links intended for human consumption would also have a relative hint so that my browser could turn the cryptographic identifier back into a human readable identifier in my own namespace.

jeffparsons• 4 years ago

I guess I'm not adding much to your comment, but I just wanted to remark that this exact thing has been bugging me for ages: people talk about these blockchain projects as being "decentralized" simply because _execution of the algorithm_ is distributed across lots of (mutually distrusting) nodes. This perspective confuses me, because it doesn't seem to be a particularly meaningful distinction when there is still one consensus truth in the network.

In the case of a naming system, as you point out, that means there is still one finite, scarce namespace in the system, just as there is in DNS.

More generally, there is still a central authority pushing out code, and so unless you want to hard-fork, what does the "decentralization" in a blockchain actually buy you? Does it do anything more for you in practical terms than protecting you from a central server suffering an outage?

I spent some time a while back coming up with ideas for an alternative to blockchains that would pursue many of the same goals, starting from first principles ("what the heck even _is_ money?"). I have no idea whether any of my ideas would have actually worked in practice, but they at least ticked the "decentralized" box in a way that I found a lot more meaningful than merely "decentralized execution of a centrally-defined algorithm to determine a single global consensus state". (I stopped working on them not because they didn't show promise, but because I became convinced that they would be a net negative for the world if they were successful; I lost the desire to find a working design.)

some-username• 4 years ago

You might want to have a look at the GNS (GNU Name System): https://www.gnunet.org/en/gns.html https://datatracker.ietf.org/doc/id/draft-schanzen-gns-01.ht... To my understanding, this enables what you are looking for.

sanxiyn• 4 years ago

This has been noticed quickly. Aaron Swartz wrote "Squaring the Triangle" in 2011: http://www.aaronsw.com/weblog/squarezooko.

api• 4 years ago

Proof of work gets progressively centralized by industrial economies of scale. Proof of stake will lead to a plutocracy controlling the chain.

kelthuzad3• 4 years ago

Not all PoW mining were created equal, monero for instance uses a pow algorithm that was designed to be resistant to ASIC mining.

edflsafoiewq• 4 years ago

Here's another triangle

Human-meaningful: Meaningful and memorable names.

Unique: The referent of a name is unique and immutable.

Intergenerational: The first generation of users can't take up all the good names.

cowtools2• 4 years ago

I think the intergenerational aspect is intertwined with the "centralized" aspect if zooko's triangle, because traditionally in DNS you have some central authority (ICANN, IETF, registrar) that is expected to "solve" that somehow by deciding who should be allowed to own certain domains.

eternityforest3• 4 years ago

This is another good case for just tossing the whole unique name thing entirely and going with key hashes and freely chosen nonunique names.

qwerty456127• 4 years ago

IMHO there should be no good or bad names and no choice which do you take. I loved ICQ so much because it didn't require me to invent a nickname nor to struggle spelling it aloud to people from different countries. Just a reasonably short number - that was amazing.

rapnie• 4 years ago

There's also a nice short paper [0] sent to Rebooting the Web of Trust by Christine Lemmer-Webber, Mark S. Miller et al. The Spritely Institute [1] is elaborating on this [2]

[0] https://github.com/cwebber/rebooting-the-web-of-trust-spring...

[1] https://spritely.institute

[2] https://spritelyproject.org/news/petname-systems.html

eternityforest3• 4 years ago

Human meaningful names aren't great. They basically require that names have a cost. Either DNS or blockchain, and either way you may have trouble on isolated networks.

I think a better solution is to just leave meaningful names to centralized DNS. It's a perfect fit. Amazon is centralized already.

Private individuals don't commonly need them. Look at phone numbers. We are fine with local assigned names for the most part.

What we need is a way to create a URL that embeds its SSL fingerprint and is independent of any authority. Self hosted stuff can use that because there's generally a safe bootstrap method to communicate that name.

And even with unsafe bootstrapping like non unique search, you are still not any worse off than you are with Facebook using the same method. Probably better, in fact, if you are using local-only nonunique search over Bluetooth instead of something anyone anywhere can fake.

Meaningful names can even give a false sense of security and lead to typosquatting. The advantage of not needing a blockchain or consensus seems bigger than the advantage of names.

layer8• 4 years ago

Human meaningful names are useful when used in configuration files and the like, because otherwise you have no way to recognize what host is being referenced, and to easily communicate with others about it. I would maybe prefer the term “human-recognizable names” for that reason.

eternityforest3• 4 years ago

In a local context they're great, but it seems like Zooko only applies to global DNSlike systems.

Zooko's Triangle

Related Stories