8 ms·

Unfortunately, there are a number of things at play here that overall results in progress towards anyone being able to be "their own YouTube" very difficult. *

by dkh 6y ago

Unfortunately, there are a number of things at play here that overall results in progress towards anyone being able to be "their own YouTube" very difficult. * Storage costs have decreased over the years and will continue to do so, but video quality has increased during this time, as well as the number of delivery formats you need to create. For each video uploaded, you likely need to encode and store 5+ different variations. * Bandwidth is not cheap, and while it may be lower than the cost of storage initially, for any chance of survival you'll eventually need some popular content, at which point its cost of delivery will exceed that of storage * To prevent the video from buffering and provide an experience that is tolerable to the average user, you need to have all of the above replicated on a dozen edge nodes around the world, and serve a user the video they've requested from the one nearest to them. This is a situation where the latency differences here really can make or break the entire thing There are numerous ways to try to optimize these things, or filter out costly/undesirable content, or prioritize only things that people are actually watching, or whatever, but few that come without consequences or that will cause bigger problems down the line.

dvfjsdhgfv 6y ago
Fortunately, all these can be reasonably dealt with for niche videos (say, <10k plays by month). You could keep a collection of a few thousand of such videos on one Hetzner box for €20-30 a month (skipping wouldn't be smooth, put normal playback should be). This would make sense for small orgs though, not individuals.
- barrkel 6y ago
  What's needed is some kind of almost insurance scheme so that the long tail of content can be served up, but if a video goes viral, it doesn't become completely unavailable. Needs care with adverse risk selection, because freeloaders who want cheap video bandwidth for god knows what will crowd around.
  - anderspitman 6y ago
    That's an interesting idea. Could default to slow performance served directly off your laptop, and if it goes viral the network automatically moves it to a CDN at no extra cost to you. As an extra bonus, whatever small fee you pay to buy in isn't a complete waste of money, because presumably you get value from the viral content of others, ie as a consumer of their content.
ggoo 6y ago
I think this is a much more solvable problem than people think, it just requires not considering everything in such a centralized manner. We don't have to build "their own YouTube", content can be shared peer to peer - the torrent community has solved many of these issues already.
- nine_k 6y ago
  Most importantly, the audience should expect and accept a delay. Youtube goes to great lengths to make videos play instantly, including running huge local caches near large traffic exchange points. The audience is conditioned to see a clip start in under a second, faster for popular clips. Anything that runs off distributed (not replicated) storage, let alone peer-to-peer links, will inevitably be slower to start. It may take 10 seconds, or 30 seconds, for a clip to begin playing. This alone can kill such solutions for the mass audience, hungry for instant gratification.
  - Taek 6y ago
    Sia is able to load files over a peer-to-peer network in under 200ms, and has streaming solutions built in which allow for super smooth playback. A lot of people have this bias that p2p networks cannot be fast, but those days are behind us.
  - reedwolf 6y ago
    >Most importantly, the audience should expect and accept a delay Demanding that people accept the shortcomings of your product/service as a fact of the Universe isn't going to get you far. >hungry for instant gratification How many hours of video content are uploaded to the web every second? Life is too short to wait around to see if your video is worth watching in full.
    - nine_k 6y ago
      Thank you for a nice illustration.
  - asdff 6y ago
djsumdog 6y ago
What about using PeerTube and other technologies that utilize webtorrent in combination with WebRTC? People should help others stream videos while watching them, rather than let a sole provider do all the heavy lifting.
- lovvtide 6y ago
  https://satellite.earth/ https://satellite.earth/ is doing exactly that. Using WebTorrent (over WebRTC) for the distribution layer and Ethereum as a decentralized an auth layer so people can digitally sign everything they upload and prove authorship of the data they're sharing with peers. @feross
  - MayeulC 6y ago
    Peertube has been doing exactly that as well. I don't really see what Etherum brings to the table. You can have signatures without the weight of a blockchain. Actually not relying on one is more portable across isolated networks, and requires way less computing resources. Plus, blockchains are usually a privacy issue. I'm highly interested in the topics of decentralized (federated or distributed) platforms and identities, but I don't see what a single blockchain brings to the table? I could see profiles-as-blockchains, though, or instances-as-blockchains. But the rationale isn't clearly explained in the website, so it feels more of a buzzword and less of something that efficiently solves an issue (content discovery and global indexing, I guess). Lastly, would you mind stating your affiliation? Thanks :)
    - dkh 6y ago
      Just as an example of how crypto can help solve some problems in this arena see VideoCoin's proof system: (https://medium.com/videocoin/proof-system-for-videocoin-storage-mining-4c8f8b79bb90 https://medium.com/videocoin/proof-system-for-videocoin-stor... and https://medium.com/videocoin/proof-of-transcoding-as-a-new-consensus-mechanism-ca9ccf3fa168 https://medium.com/videocoin/proof-of-transcoding-as-a-new-c...) They are attempting to decentralize and distribute video encode/delivery, and a big piece of that is a poof of work implementation to incentivize and reward reliable nodes. Just one example, but there are others, and a lot of progress has been made to reduce the strain caused by throwing crypto into something that fundamentally needs to be very fast and lightweight to be usable. Edit: Corrected duplicate links
ehnto 6y ago
I think looking for a "YouTube Killer" is the wrong direction anyway, unless you're just looking for another large corporation to take over video dominance. What I would personally love is an aggregator, or many aggregators, that act like YouTube, but are actually a distributed network of video sources. The key to the concept is thinking of a source like a library of content, and the individual uploader manages their own source. You could consider it a channel, and then that uploader can syndicate their source/channel to any aggregator they want. Sources can be private/public, sources can be tagged, discoverable, and searchable. Another key point is that uploaders need not host things themselves, SaaS products will quickly fill that role. a) Anyone can start an aggregator and immediately have content, lowering the bar for entry. b) A niche aggregator can pick or have submitted to them video library sources that match their niche, and can curate them just like any other video hosting site. c) A cottage industry for video hosting can utilize an "average user" friendly UI to allow for anyone to create a source and submit it to these aggregators. d) Allows free information to flow freely f) Allows content creators full control over deals with aggregators on how they'll be reimbursed for access to their source. You can imagine YouTube accepting sources with an ad revenue share model just like they have now. Some aggregators will choose to "cache and redistribute" content from sources to allow for consistent UX to their users, but that's their decision.
- vkou 6y ago
  You're going to get a flood of porn, pirated content, illegal-for-other-reasons content, and, of course, nazis who have been banned from all the other platforms. Which will, of course, drive all the normies away. (Well, okay, all the pirated content might not, but then you'll also include 'all the media companies' and 'the government' as your list of enemies.) And if you add federated moderation (Like Mastodon did), you're going to quickly understand why federated moderation is a labour-sucking, drama-ridden, centralized-censorship, tire pyre.[1] (Like Mastodon did.) [1] Mastodon instance administrators blacklist instances that produce problem content and problem users.[2] (Child pornography, nazis, furries, etc.) These ban-lists are centralized, because nobody has time to investigate every single other instance in the network. I hope you trust the person running those centralized lists! [2] Oh, and if a troll group does not like a particular person, they can follow him around, shitting on every instance that he posts in. How do instance operators deal with this sort of thing? By banning[3] the target of the trolls, because then they can stop wasting time playing whack-a-mole. [3] https://www.theverge.com/2018/8/31/17801404/mastodon-harassment-wil-wheaton-mobs-twitter https://www.theverge.com/2018/8/31/17801404/mastodon-harassm...
Taek 6y ago
Working on an infrastructure layer project to address this called Skynet - https://siasky.net https://siasky.net Data storage costs at around $1/TB/Mo raw ($3 for 3x redundancy), bandwidth costs at around $1 / TB egress. Pricing comes from an open marketplace, and we do a lot of sophisticated stuff to make sure it's high reliability. We've got video buffering solved as well. The latest builds of Sia (decentralized storage network) can retrieve files at about 700mbps with a time-to-first-byte of under 200ms. We haven't fully rolled this upgrade out to siasky.net yet, but it should be there in the next 1-2 weeks. The decentralized cloud is a lot closer than people realize, the infrastructure is finally reaching a point of being competitive.
- dkh 6y ago
  I've been cautiously following projects like this, and will add this one to the list.
  - anderspitman 6y ago
    Do you happen to have a list you can share?
- hirako2000 6y ago
  I'm sorry, egress is far more expensive than that.
  - Taek 6y ago
    https://siastats.info/storage_pricing https://siastats.info/storage_pricing Current rates on the network are $0.70 per TB, cloud to user. Many of the nodes are serving bandwidth from unlimited connections, either at home or from cloud providers like Hetzner.
chaostheory 6y ago
Let's not forget the issues like SESTA & FOSTA
MayeulC 6y ago
Peertube neatly solves part of those issues. * Storage costs are likely paid for by the uploader. Most regular uploader do (or should) keep at least a local copy of their work. Plus the original materials. So the weight of the video is likely not that significant. * With webtorrent, the more a video is watched, the more it is shared. If you want to handle the worst case though, you'd need enough bandwidth to serve a couple times every file over simultaneously. That gets worse the more videos you have, but simultaneous watching of every one of your videos is unlikely, I think, over a certain number. It would be nice to get some stats on this. * That's not really needed, is it? Latency isn't likely to be big enough to cause issues, and I would think available bandwidth doesn't change that much with distance. With a couple of local torrent peers, that also gets better. Now, this isn't perfect, but there are ways this can be improved: * Peertube intances can mirror content to provide at least one peer (and a backup HTTP source I think). It would be nice if subscribers could choose to support their favorite channels by asking their instance to mirror that content (with a lower priority than local content, of course). That could more or less solve the local CDN issue, if "peertubers" participate in content exchange to reduce downtime and improve experience. A form of insurance, if you want. * I wish scalable video (SVC, or adaptative video bitrate, however you want to call it) was more supported by ffmpeg & container formats. Using a single file for the various qualities would be a natural fit for Webtorrent, as it would allow HQ watchers feed LQ watchers. * Multicast would be a great thing to have on the Internet. Maybe some future/alternative IP stack implementations could bring that back? Yggdrasil is promising, for instance. If nodes could subscribe to a given content, I guess it could be delivered across the network, splitting where it should.
syshum 6y ago
YouTube's (and most social media) value is not in its storage and bandwidth. it is in Discoverability. That is the problem the tech community needs to solve, not storage or bandwith, ways to distribute that is really a solved issue in several ways. The problem for creators to move off of YouTube is the discoverability for an audience.
- dkh 6y ago
  It is definitely true that an enormous part of YouTube's value is its ubiquity and the discoverability of content there. Absolutely. But the immense technical and financial challenges are a huge part of why it's such a Herculean task to build an alternative and survive long enough to provide value.
asdff 6y ago
That's the CDN model, but what about the peer to peer model? I can stream a video over bittorrent with just one good seeder.
- giantrobot 6y ago
  You might be able to stream over BitTorrent with one good seeder. It doesn't take many connections to saturate that seeder's upstream capability. Once a torrent's total seed upstream capacity is exceeded clients' playback will stall until they can buffer enough video to restart playback. It takes a lot of upstream bandwidth to stream even just passable quality video. Between shitty upstream bandwidth for residential Internet and swarm participants limiting their seed ratios or further limiting upstream bandwidth scaling torrent streaming is unreliable at best and impractical at worst.
superkuh 6y ago
Peer to peer video streaming on top of bittorrent has already existed multiple times and it has worked well at scale. It's just that there wasn't centralized control and monetization of it so each time it was shut down by the legal authorities. The problem is not technical. It's legal.
the8472 6y ago
> * Bandwidth is not cheap, and while it may be lower than the cost of storage initially, for any chance of survival you'll eventually need some popular content, at which point its cost of delivery will exceed that of storage If only IP-SSM worked on the open internet. Even some sort of best-effort routing would be a start.

by dkh 6y ago

dvfjsdhgfv 6y ago

Fortunately, all these can be reasonably dealt with for niche videos (say, <10k plays by month). You could keep a collection of a few thousand of such videos on one Hetzner box for €20-30 a month (skipping wouldn't be smooth, put normal playback should be). This would make sense for small orgs though, not individuals.

barrkel 6y ago
What's needed is some kind of almost insurance scheme so that the long tail of content can be served up, but if a video goes viral, it doesn't become completely unavailable. Needs care with adverse risk selection, because freeloaders who want cheap video bandwidth for god knows what will crowd around.
- anderspitman 6y ago
  That's an interesting idea. Could default to slow performance served directly off your laptop, and if it goes viral the network automatically moves it to a CDN at no extra cost to you. As an extra bonus, whatever small fee you pay to buy in isn't a complete waste of money, because presumably you get value from the viral content of others, ie as a consumer of their content.

ggoo 6y ago

I think this is a much more solvable problem than people think, it just requires not considering everything in such a centralized manner. We don't have to build "their own YouTube", content can be shared peer to peer - the torrent community has solved many of these issues already.

nine_k 6y ago
Most importantly, the audience should expect and accept a delay. Youtube goes to great lengths to make videos play instantly, including running huge local caches near large traffic exchange points. The audience is conditioned to see a clip start in under a second, faster for popular clips. Anything that runs off distributed (not replicated) storage, let alone peer-to-peer links, will inevitably be slower to start. It may take 10 seconds, or 30 seconds, for a clip to begin playing. This alone can kill such solutions for the mass audience, hungry for instant gratification.
- Taek 6y ago
  Sia is able to load files over a peer-to-peer network in under 200ms, and has streaming solutions built in which allow for super smooth playback. A lot of people have this bias that p2p networks cannot be fast, but those days are behind us.
- reedwolf 6y ago
  >Most importantly, the audience should expect and accept a delay Demanding that people accept the shortcomings of your product/service as a fact of the Universe isn't going to get you far. >hungry for instant gratification How many hours of video content are uploaded to the web every second? Life is too short to wait around to see if your video is worth watching in full.
  - nine_k 6y ago
    Thank you for a nice illustration.
- asdff 6y ago

djsumdog 6y ago

What about using PeerTube and other technologies that utilize webtorrent in combination with WebRTC? People should help others stream videos while watching them, rather than let a sole provider do all the heavy lifting.

lovvtide 6y ago
https://satellite.earth/ https://satellite.earth/ is doing exactly that. Using WebTorrent (over WebRTC) for the distribution layer and Ethereum as a decentralized an auth layer so people can digitally sign everything they upload and prove authorship of the data they're sharing with peers. @feross
- MayeulC 6y ago
  Peertube has been doing exactly that as well. I don't really see what Etherum brings to the table. You can have signatures without the weight of a blockchain. Actually not relying on one is more portable across isolated networks, and requires way less computing resources. Plus, blockchains are usually a privacy issue. I'm highly interested in the topics of decentralized (federated or distributed) platforms and identities, but I don't see what a single blockchain brings to the table? I could see profiles-as-blockchains, though, or instances-as-blockchains. But the rationale isn't clearly explained in the website, so it feels more of a buzzword and less of something that efficiently solves an issue (content discovery and global indexing, I guess). Lastly, would you mind stating your affiliation? Thanks :)
  - dkh 6y ago
    Just as an example of how crypto can help solve some problems in this arena see VideoCoin's proof system: (https://medium.com/videocoin/proof-system-for-videocoin-storage-mining-4c8f8b79bb90 https://medium.com/videocoin/proof-system-for-videocoin-stor... and https://medium.com/videocoin/proof-of-transcoding-as-a-new-consensus-mechanism-ca9ccf3fa168 https://medium.com/videocoin/proof-of-transcoding-as-a-new-c...) They are attempting to decentralize and distribute video encode/delivery, and a big piece of that is a poof of work implementation to incentivize and reward reliable nodes. Just one example, but there are others, and a lot of progress has been made to reduce the strain caused by throwing crypto into something that fundamentally needs to be very fast and lightweight to be usable. Edit: Corrected duplicate links

ehnto 6y ago

I think looking for a "YouTube Killer" is the wrong direction anyway, unless you're just looking for another large corporation to take over video dominance. What I would personally love is an aggregator, or many aggregators, that act like YouTube, but are actually a distributed network of video sources. The key to the concept is thinking of a source like a library of content, and the individual uploader manages their own source. You could consider it a channel, and then that uploader can syndicate their source/channel to any aggregator they want. Sources can be private/public, sources can be tagged, discoverable, and searchable. Another key point is that uploaders need not host things themselves, SaaS products will quickly fill that role. a) Anyone can start an aggregator and immediately have content, lowering the bar for entry. b) A niche aggregator can pick or have submitted to them video library sources that match their niche, and can curate them just like any other video hosting site. c) A cottage industry for video hosting can utilize an "average user" friendly UI to allow for anyone to create a source and submit it to these aggregators. d) Allows free information to flow freely f) Allows content creators full control over deals with aggregators on how they'll be reimbursed for access to their source. You can imagine YouTube accepting sources with an ad revenue share model just like they have now. Some aggregators will choose to "cache and redistribute" content from sources to allow for consistent UX to their users, but that's their decision.

vkou 6y ago
You're going to get a flood of porn, pirated content, illegal-for-other-reasons content, and, of course, nazis who have been banned from all the other platforms. Which will, of course, drive all the normies away. (Well, okay, all the pirated content might not, but then you'll also include 'all the media companies' and 'the government' as your list of enemies.) And if you add federated moderation (Like Mastodon did), you're going to quickly understand why federated moderation is a labour-sucking, drama-ridden, centralized-censorship, tire pyre.[1] (Like Mastodon did.) [1] Mastodon instance administrators blacklist instances that produce problem content and problem users.[2] (Child pornography, nazis, furries, etc.) These ban-lists are centralized, because nobody has time to investigate every single other instance in the network. I hope you trust the person running those centralized lists! [2] Oh, and if a troll group does not like a particular person, they can follow him around, shitting on every instance that he posts in. How do instance operators deal with this sort of thing? By banning[3] the target of the trolls, because then they can stop wasting time playing whack-a-mole. [3] https://www.theverge.com/2018/8/31/17801404/mastodon-harassment-wil-wheaton-mobs-twitter https://www.theverge.com/2018/8/31/17801404/mastodon-harassm...

Taek 6y ago

Working on an infrastructure layer project to address this called Skynet - https://siasky.net https://siasky.net Data storage costs at around $1/TB/Mo raw ($3 for 3x redundancy), bandwidth costs at around $1 / TB egress. Pricing comes from an open marketplace, and we do a lot of sophisticated stuff to make sure it's high reliability. We've got video buffering solved as well. The latest builds of Sia (decentralized storage network) can retrieve files at about 700mbps with a time-to-first-byte of under 200ms. We haven't fully rolled this upgrade out to siasky.net yet, but it should be there in the next 1-2 weeks. The decentralized cloud is a lot closer than people realize, the infrastructure is finally reaching a point of being competitive.

dkh 6y ago
I've been cautiously following projects like this, and will add this one to the list.
- anderspitman 6y ago
  Do you happen to have a list you can share?
hirako2000 6y ago
I'm sorry, egress is far more expensive than that.
- Taek 6y ago
  https://siastats.info/storage_pricing https://siastats.info/storage_pricing Current rates on the network are $0.70 per TB, cloud to user. Many of the nodes are serving bandwidth from unlimited connections, either at home or from cloud providers like Hetzner.

chaostheory 6y ago

Let's not forget the issues like SESTA & FOSTA

MayeulC 6y ago

Peertube neatly solves part of those issues. * Storage costs are likely paid for by the uploader. Most regular uploader do (or should) keep at least a local copy of their work. Plus the original materials. So the weight of the video is likely not that significant. * With webtorrent, the more a video is watched, the more it is shared. If you want to handle the worst case though, you'd need enough bandwidth to serve a couple times every file over simultaneously. That gets worse the more videos you have, but simultaneous watching of every one of your videos is unlikely, I think, over a certain number. It would be nice to get some stats on this. * That's not really needed, is it? Latency isn't likely to be big enough to cause issues, and I would think available bandwidth doesn't change that much with distance. With a couple of local torrent peers, that also gets better. Now, this isn't perfect, but there are ways this can be improved: * Peertube intances can mirror content to provide at least one peer (and a backup HTTP source I think). It would be nice if subscribers could choose to support their favorite channels by asking their instance to mirror that content (with a lower priority than local content, of course). That could more or less solve the local CDN issue, if "peertubers" participate in content exchange to reduce downtime and improve experience. A form of insurance, if you want. * I wish scalable video (SVC, or adaptative video bitrate, however you want to call it) was more supported by ffmpeg & container formats. Using a single file for the various qualities would be a natural fit for Webtorrent, as it would allow HQ watchers feed LQ watchers. * Multicast would be a great thing to have on the Internet. Maybe some future/alternative IP stack implementations could bring that back? Yggdrasil is promising, for instance. If nodes could subscribe to a given content, I guess it could be delivered across the network, splitting where it should.

syshum 6y ago

YouTube's (and most social media) value is not in its storage and bandwidth. it is in Discoverability. That is the problem the tech community needs to solve, not storage or bandwith, ways to distribute that is really a solved issue in several ways. The problem for creators to move off of YouTube is the discoverability for an audience.

dkh 6y ago
It is definitely true that an enormous part of YouTube's value is its ubiquity and the discoverability of content there. Absolutely. But the immense technical and financial challenges are a huge part of why it's such a Herculean task to build an alternative and survive long enough to provide value.

asdff 6y ago

That's the CDN model, but what about the peer to peer model? I can stream a video over bittorrent with just one good seeder.

giantrobot 6y ago
You might be able to stream over BitTorrent with one good seeder. It doesn't take many connections to saturate that seeder's upstream capability. Once a torrent's total seed upstream capacity is exceeded clients' playback will stall until they can buffer enough video to restart playback. It takes a lot of upstream bandwidth to stream even just passable quality video. Between shitty upstream bandwidth for residential Internet and swarm participants limiting their seed ratios or further limiting upstream bandwidth scaling torrent streaming is unreliable at best and impractical at worst.

superkuh 6y ago

Peer to peer video streaming on top of bittorrent has already existed multiple times and it has worked well at scale. It's just that there wasn't centralized control and monetization of it so each time it was shut down by the legal authorities. The problem is not technical. It's legal.

the8472 6y ago

> * Bandwidth is not cheap, and while it may be lower than the cost of storage initially, for any chance of survival you'll eventually need some popular content, at which point its cost of delivery will exceed that of storage If only IP-SSM worked on the open internet. Even some sort of best-effort routing would be a start.