Comment by jasode | Hacker News Reader

Comment by jasode | Hacker News Reader

Parent Full thread

Because the laws of physics (speed of light) makes a decentralized discovery service slower which means it's extremely inconvenient for typical users.

For example, go to youtube.com and see how fast it loads the main landing page with 20+ suggested video thumbnails. On my computer, it loads in 1136 milliseconds. The same centralized-system's speed advantage happens for video search results such as "3d printing".[1]

A decentralized system with DHT[2] or other peer-2-peer scheme cannot match that speed. Think about trying to query dozens (or potentially thousands) of p2p nodes to return a page of video thumbnails. If you try to "solve" the slow performance by "caching" results, you still have to cache it on a server somewhere. That "cache server" is an example of feature-creep towards centralization.

People will put up with decentralized's inconveniences for illegal things like bittorrent pirated movies. For the vast majority of mainstream content, a centralized service is too attractive for billions of web users.

[1] https://www.youtube.com/results?search_query=3d+printing

[2] https://en.wikipedia.org/wiki/Distributed_hash_table

The claim you make (that the speed of light) is the main limiting factor is incredibly undersubstantiated. Firstly, it is incredibly likely that the same query within you tube is handled by as many computers as it would have been in your DHT.

The difference between youtube and your DHT is that youtube doesn't need to do constant connection re-establishment. In a DHT -- especially one used for short-lived queries such as these -- the establishing of short connections can add a lot of overhead. This is a challenge mainly because of the protocol stocks in common usage on the network, not really a property of distributed systems in and of themselves.

>same query within you tube is handled by as many computers as it would have been in your DHT.

For the poster I responded to, and the topic of this thread (Peertube), decentralized means WAN (wide area network) and not LAN (local area network).

(I noticed you substituted "decentralized" with "distributed" which muddies up the discussion. The poster asking the question and my response used "decentralized".)

>because of the protocol stocks[sic] in common usage on the network, not really a property of distributed systems in and of themselves.

The overhead of network latency and multiple round trips is absolutely a property of decentralized systems. It's technically impossible to invent a protocol that can query thousands of home-based nodes to return a result that's as fast as a centralized system such as Youtube.

For the purposes of this particular thread, the "distributed network" means something like PeerTube with home nodes over ISP networks. We're not talking about the "distributed network" of 10,000 computers within Google and Facebook datacenters.

[To downvoters, please point out the technical flaw or show a protocol stack that removes ISP latency and round trips that makes it equivalent network performance to Youtube as tathougies suggests.]

> It's technically impossible to invent a protocol that can query thousands of home-based nodes to return a result that's as fast as a centralized system such as Youtube.

Yeah, but it's possible to crawl and cache all metadata, everything except the actual video and audio, centrally, and search it from there. Sure, that means you can't find a video 5 seconds after it was uploaded, but for practical purposes, it's not hard to imagine a bunch of solutions that would be good enough, especially considering search isn't that great on YT to begin with.

Why can't I search among subscribed channels and filter and order by 20 criteria? Because search on YouTube isn't that good, it's not even trying to be, just like on Facebook for example. Being quick at doing something crummy is still kinda crummy.

I always considered my favourite system to be one where the discovery system was centralized, but the actual content itself and the ability to produce it was decentralized. So, for instance, you'd have "PeerTube" as a protocol for uploading your videos to be discovered, then you'd use something like "Youtube" to view videos from multiple peers. This prevents a massive amount of lockin as happens with the current services, but also enables all the QoS you expect from modern websites.

Torrent works exactly like this, doesn't it? All it needs is WebTorrent to become more popular so you can stream videos directly from the browser.

>That "cache server" is an example of feature-creep towards centralization.

That's a tiresome slippery slope argument. "Better not install that proxy server or soon everyone will host all of their content on it instead of on their own servers!"

ToR is decentralized, it has servers. Bitorrent is decentralized, it has servers. Etc etc.

Reply on news.ycombinator.com