Hacker News Reader

Comment by nemothekid | Hacker News Reader

>I believe that we are going to move away from a lot of traditional ways of doing things pretty soon. For one, obviously centralized servers have to go

This is a pretty bold statement, and I'd like to hear your reasoning for it.

I've always seen that centralized is better than decenteralized, just because its easier. Which is why when we see github go down, some teams can't continue to work even though git itself is decentralized. One click solutions, and decentralized applications tend to not go hand in hand.

I'll agree, if we want to maintain privacy these things will have to be decentralized. However I have a hard time making the case that decentralization will become the defacto standard. Even the web has a psuedo-central hub - DNS.

View on HN

I would say centralised is easier but usually worse instead of better. Most people use some centralised middleman out of convenience. Decentralisation is universally better IMO because there is no single point of failure and no choke point to nickel-and-dime users to death.

> some teams can't continue to work even though git is decentralized.

I think you mean some teams won't continue to work because they don't know how to use git. If the de facto center goes down, a team is still able to use git to share changes with each other.

And IPv4/6 is decentralised, DNS makes it convenient for regular people to use. I would say DNS is decentralised as well, just the allocation of domain names comes from a few sources.

A nitpick, but decentralized & distributed systems do not guarantee "no single point of failure". For example, Skype switched from pure p2p because of a software bug which affected the whole network (ie the client software was the single point of failure)...

That's not true. The whole network was effected because the centralization point they chose had a client bug and since the whole network had to use these nodes it took them down.

To take down a decentralized network with a bug you'd have to have every user on the exact same version and no way for them to simply role back to the previous version. I.e. nearly impossible.

That's also not true - all you need is to bring down a high enough percentage of nodes, which will effectively take down the network through capacity constraints.

Combine this with auto-updating clients and you've got a crashing software bug that brings down the whole network.

Agreed however that the centralisation point effectively still exists thanks to automated updates.

>all you need is to bring down a high enough percentage of nodes, which will effectively take down the network through capacity constraints

On a fully distributed network, that's pretty hard to do. DDOS works by having tons of nodes to attack with. That won't work very well when your target has as many or more nodes than you do.

Taking down a centralized target is vastly easier.

>Combine this with auto-updating clients and you've got a crashing software bug that brings down the whole network.

Only if I can't role my client back nor turn off auto-update.

Reply on news.ycombinator.com