I wonder if such a system really even needs domains anymore. Would it be possible to scrap domains altogether and use IPs only?
The link structure of the web is almost completely based on domain urls, but I wonder if there's not some way to work around that in a DNS-less/P2P system.
Many common services (HTTP, SMTP, IMAP, POP3, and especially DNS, if you think about it) are provided by daemons that don't really care much about the domain name of the machine they're running on. For example, you can configure your web server to deliver pages for www.example.com, and it will, as long as that domain is in the HOST: header of the request. No DNS is required, you just point the request at the web server's IP address.
The obvious problem is that, to my knowledge, you can't embed a HOST: header in a URL to fetch that resource from an arbitrary IP address (something like http://HOST:[email protected]/).
Like HTTP, SMTP servers will gladly accept messages for domains it is configured to handle. But it also depends on DNS to get MX (or A) records to deliver to domains it doesn't handle. It's trivial to support email addresses that use IPs instead of domains (like [email protected]), but such addresses are less portable than using domains and also create conflicts because two users cannot have the same name, even if they operate in different realms. Besides, they're butt-ugly and harder to remember than domains.
tl;dr: Using IPs only creates problems and DNS is a HUGE part of the solution. Any replacement will have to solve the same problems.
DNS, URIs, and application-level protocols such as HTTP and SMTP work together, but that doesn't mean they are the same beast. The reason for the existence of URIs is to provide identifiers for resources. DNS makes these human-readable. Applications in turn use these facilities.
When the user types http://www.example.com into the address bar, it's the web browser that figures out what to do next. Which is: realize it needs to to a HTTP request. Where to? Not an IP, ask DNS. Now connect to the IP address. But an HTTP server can host multiple domains, so include the host name in the request (that's the Host header). The web server then looks in its configuration, and sends the right page back. Note that the HTTP headers are specific to the application protocol, and are irrelevant both at the DNS and URL level. It just happens to be the same string :)
Domains make IPs human-readable and human-rememberable, and allow multiple servers to identify as the same site (ie, for human and CA purposes), and allow them to move. Somewhere you need a translation and connection system.
You could most definitely change your links to go to IP addresses. But they're more fragile.
Server IP addresses change every time you switch to a different provider.
Ironically the DNS actually provides a layer of resilience against being forced offline by targeting your provider, but that moves the vulnerability up one level, now you rely on the DNS to do its thing, and it too is under the control of a single institution.
The bar is a bit higher than a C&D to your provider but apparently not nearly high enough.
The link structure of the web is almost completely based on domain urls, but I wonder if there's not some way to work around that in a DNS-less/P2P system.