The downsides are that it is too centralized - it isn't difficult to imagine that a government agency would want to sinkhole silkroad.tor from the default registry.
With an alternate registry, you have the balance between knowing enough about the directory provider so that you can trust them, but not enough known about then where they are open to legal recourse.
ie. i'd trust a registry from riseup or duckduckgo, but that same registry is likely going to be the target of legal and hacking attempts. Likewise any provider who is sufficiently protected from those threats likely isn't well-known enough to be trustworthy.
One of the benefits of the existing names is that they also authenticate the site (assuming you check it correctly, usually out of band from a trusted source like a directory or search engine) - this part can be replaced with certificates and an issuance model that can be identical to what LetsEncrypt does
In terms of hosting the directory - that almost has to be decentralized using a p2p network. Similar to namecoin. Namecoin also solves the issue of distributing names and typosquatting - and it could be adapted to auction names.
It is only centralized for users with the default install, who never go into their address book.
I think the real achievement of I2P's name system is that _they have made it easy for users to understand_, and the tight integration in the UX is the main differentiator I see between I2P's approach and any of the approaches in this blog post.
While I think Namecoin sounds cool and all, I really hope Tor considers a simpler approach. I think it's a mistake for us to make this into a technical problem, when it's a UX problem. We're never going to get 100% secure names in a trustless environment, so why not focus on making the default pretty secure, and making the system understandable and useable?
this seems a good example of "fruitful surface area" for security vulns. another unrelated design decision is for Tor to reuse a single route which can be identified with relatively high accuracy via traffic analysis.
Also, its important to note that I2P's addresses (which are usually seen as something.i2p), can also be used directly in their 'hash' form like Tor, if you don't trust the jump/address book services.
Another thing I forgot to mention: if a name is unknown, your local HTTP proxy will ask you if you want to try to lookup the name from one of several popular "jump services", which are the same services providing address book subscriptions. This is where jump links come from most of the time.
- Users have a local "Address Book" which maps friendly names (e.g. forum.i2p) to I2P destination keys.
- There are well-known I2P hidden services providing address book subscriptions. The default install includes a subscription that is maintained (and signed!) by the project maintainers.
- The address book makes it clear where names are coming from. So if you decide to un-trust one of your existing subscriptions, you can still keep addresses you added yourself, etc.
- New sites submit their name & key to popular address book services, but there is an additional trick you can use. Pass people a link like http://mysite.i2p/?i2paddresshelper=<key>. This is known as a "jump" link. Your local HTTP proxy will take you to a page asking if you would like to add the name to your address book, or if you'd simply like to keep the name for this session but not save it.
The whole system is easy to use while staying flexible, secure, and in the user's control. I'm curious if the Tor team has tried out I2P or considered a system like this.
If it's not making sense to you from my summary, I highly encourage you to get I2P and try it out!
[0] https://geti2p.net/en/ [1] https://geti2p.net/en/docs/naming