Tim Berners-Lee: We need to re-decentralise the web

Web apps are a bit of an issue, but unhosted web apps are not more of an issue than applications (as opposed to utilities composable in the shell) in general - which is a tremendous improvement.

While I agree that TCP/IP is decentralized, unfortunately due to NAT and corporate firewalls, vast swaths of users have been relegated to second class citizens who can only act as data consumers. Unfortunately this same misguided approach to security, or as I call it suckerity, has been carried forward into IPv6.

There is some hope that WebRTC and other p2p technologies might alleviate the problem but then the task of punching through firewalls with techniques like STUN falls on developers. I'm not seeing an industry-led campaign to standardize a true p2p protocol based on something like UDP, because it would undermine the profits of data producers.

I could go into how the proper layering of TCP should have been IP->UDP->TCP instead of IP->TCP and a whole host of technical mistakes such as inability to set the number of checksum bits, use alternatives to TCP header compression, or even use jumbo frames over broadband, but what it all comes down to is that what we think of as “the internet” (connection-oriented reliable streams and the client/server model) is not really compatible with a distributed content-addressable connectionless internet that can work with high latencies and packet loss.

I think if this network existed, then extending HTTP to operate in a p2p fashion wouldn’t be all that complicated. Probably the way it would work is that you’d request data by hash instead of by address, and local caches would send you the pieces that match like BitTorrent. Most of the complexity will center around security, but I think Bitcoin and other networks of trust point the way to being able to both prove one’s identity and surf anonymously.

Tim Berners-Lee may not be thinking about the problem at this level but we need more people with his clout to get the ball rolling. I wasted a lot of time attempting to write a low-level stream API over UDP for networked games before zeromq or WebRTC existed and ended up failing because there are just too many factors that made it nondeterministic. It’s going to have to be a group effort and will probably require funding or at least a donation of talent from people familiar with the pitfalls to get it right. Otherwise I just don’t think a new standard (one that makes recommendations instead of solving the heart of the problem) is going to catch on.

If we were sure flash stays forever, we could just finish implementing Lightspark. With DRM that's going to be much harder.

Okay, technically HTTP does support it, but it's not standardized.

Doesn't that introduce the reverse problem though?

That makes someone else's content depend on the availability of your content - in the example of comments, the context and any intermediate replies can be lost if you stop hosting them. Leaving not a record of a conversation behind, but only indecipherable fragments.

A quick scan through the page at the link posted also seems to indicate the protocol depends a lot on content providers being well behaved. I'm assuming that somewhere in there there is handling for when content simply goes away with notification?