Hacker News Reader

Comment by zmanian | Hacker News Reader

I have a few hypothesizes about this.

1. Decentralization won in 1970 - 2005 protocol design because reliable computers were very expensive. It was economically advantageous to have a decentralized network of unreliable computers than centralized unreliable protocols. Reliable compute has now become commoditized by the cloud services. This creates a pressure for centralization.

2. Decentralized protocols have not aged well(TCP/IP, SMTP), have deep security flaws and have proven difficult to upgrade. This has created new centralization pressures to cope with the security flaws(Gmail, Cloudflare)

3. There continue to be pressures for decentralization related to both politics and security pressures. Tor and blockchain cryptocurrencies are excellent examples of these. But decentralization is unlikely to extend all the way to edge devices and instead operate in a federated fashion.

View on HN

Interesting that you describe TCP/IP as a decentralized protocol. I have the opposite opinion: That TCP/IP is the #1 forcing function for centralization in the world right now.

IP allows a maximum of 255 hops for any packet. This inherently restricts the topology of the Internet: As it stands, it can never be a world-wide decentralized mesh. Instead, you end up with large hubs and choke-points. The IP addressing scheme also makes it very difficult to have a mesh: IP addresses are assigned hierarchically. The name "Inter-Net" describes the problem directly: The Internet isn't a global network that just anyone can contribute or connect to; instead, the Internet is just a protocol for inter-connecting the world's centrally owned and operated networks. With the IP Internet forced topology, Economies-of-Scale make massive centralized services cheaper than distributed services (even if similarly massive). The obvious result: Comcast is your only ISP at home.

Disclosure: I've been working in my spare time for years on a solution, and I think an isochronous source-routed stream-based protocol is the only solution. I've got a proposed spec at IsoGrid.org

Have you seen http://www.scion-architecture.net/?

TCP/IP and Ethernet were decentralized alternatives both from a design and IP licensing point of view. But the unresolved technical debt at the transport layer is major centralizing force on the Internet right now.

I hadn't seen SCION, but just read through the FAQ and skimmed some of the Apr 2016 whitepaper. It doesn't appear to address centralization. For example from their literature: "SCION only assumes that a few top-tier ISPs in the isolation domain are trusted..." Sounds like it's trying to solve many problems, but centralization isn't one of them.

I wish them well, it's good to have lots of people working on this from all angles.

> 2. Decentralized protocols have not aged well(TCP/IP, SMTP), have deep security flaws and have proven difficult to upgrade. This has created new centralization pressures to cope with the security flaws(Gmail, Cloudflare)

Wrong, the reason centralization happened was profit. You can’t make profit with a decentralized protocol. And that’s exactly why we need more of them.

Companies trying to turn the internet into a for-profit thing are the worst possible outcome.

Reply on news.ycombinator.com