Comment by nyxxie | Hacker News Reader

Comment by nyxxie | Hacker News Reader

I've always wondered why people want to decentralize hosting of content. This, IMO, is a much harder task which leads to such restrictions as

> Please don't install PeerTube for production on a small device behind a low bandwidth connection (example: a Raspberry PI behind your ADSL link) because it could slow down the fediverse.

Why not simply decentralize the content index and let other people worry about storage for videos? When I go to Youtube, I'm there because it's a platform where I can find a large variety of entertainment without putting any effort into it, not because Google is hosting them. I'm picturing the backend here being a decentralized database of video metadata (including where and how it is hosted), and some sort of frontend where every video in the index can be viewed through a unified player.

If we're trying to give oppressed people the ability to host their controversial content, just have them host it behind Tor or over the bittorrent network and then add it to the index. Otherwise we could just let people link to videos on other video hosting sites or on their own personal websites.

Content discovery sucks nowadays. The search results bring up mostly bigger channels instead of relevant ones. Recommendations are based on view history and likes, which means you get more of the same instead of someting new. I constantly get videos I've seen already in my recommendations, these stay there for weeks until I click on them.

The major feature I used for discovery is long gone - it was the ability to add video responses, which allowed me to not only find small, niche channels with great stuff but also built a sense of community. I don't get why did they remove this,but ever since they did the site just never felt the same.

I don't know if it actually does anything (yet), but you can hide videos from results.

I mainly use it for videos I'm not interested in instead of ones that I've already seen (albeit it a while ago).

I'm not sure there's an algorithm in the world that would be able to recommend videos that aren't similar to the ones you currently watch but that you would still like. I think the best route is to just reccomend videos that are similar and rely on external sources (friend/community recommendations, Reddit, etc) for lateral movement across genres.

I can recommend OReilly's book Collective Intelligence. It shows several ways to get useful recommendation feeds from existing databases, irrespective of the similarity of content.

I've noticed a growing pattern in Youtube land. I'll watch a video of of some event, maybe a talk given by a professor. Then a week or so later I'll get a suggestion to watch the same exact talk which has been uploaded by a different user, perhaps with some minor video modification to trip up duplication filters or something. It's really lowering the quality of my suggestion feed.

At least you are getting content from other channels. It seems like all my recommendations are to videos from the last few channels I have watched. Never anything I haven't heard of before. Even most of the channels I subscribe to never show up in my feed.

It's the old exploration vs exploitation tradeoff. Most recommendation algorithms are highly optimized for exploitation and not exploration.

this is what frustrates me

I did what you said: created a video platform on RasPi + static site gen + CDN. The live streaming and everything work great. I've recorded some comparisons with IPFS / DTube and looked at BitChute / WebTorrents a bit. I remain unconvinced, meanwhile I see my system working great from first principles. Here's a show-n-tell I recorded this morning: https://www.youtube.com/watch?v=AcEVIibu1CE, and here's a self-hosted comparison of platforms: https://ispooge.com/2018-03-14-video-platforms-compared.html. The whole project is just a couple months old, could use some feedback.

The basic problem is bandwidth. If you don't decentralize hosting with something like BitTorrent, then only a few large organizations have the bandwidth to safely host videos that could go viral. A decentralized index with centralized hosting doesn't really solve this problem.

(PeerTube and WebTorrent don't fully solve this problem either, but that's because the core problem is political/economic rather than technical.)

>The basic problem is bandwidth. If you don't decentralize hosting with something like BitTorrent, then only a few large organizations have the bandwidth to safely host videos that could go viral.

Video storage would be out of scope for the index, and for good reason; other projects do it better than a completely new project not focused on decentralized file storage could be expected to.

The solution to the bandwidth problem would in fact be to just support existing storage solutions like bittorrent, ipfs, etc. So long as the video player supports it and the upload process is not complicated for non-technical users, problem solved.

While I agree in principle, I think it's because a good video streaming service is not just about the storage and the index. Adaptive delivery, such as the DASH technique [0], is a huge part of it.

[0] https://en.wikipedia.org/wiki/Dynamic_Adaptive_Streaming_ove...

Because in the very long run Silos always suck.

Content always ends up gated, restricted, metered, monitored, censored and moderated; and at some point, this will run afoul of all but the most casual users.

Reply on news.ycombinator.com