Because of the risk. A 100MB video file that "goes viral" and gets 5,000,000 downloads in a month would cost approximately $30,000 hosted on S3 (calculated using https://calculator.s3.amazonaws.com/index.html). I can't afford that. Obviously I could take the content offline if it suddenly blew up, but instead if I put it on Youtube it would cost me nothing (less in fact; they would pay me.)
Maybe at 29 I'm an old curmudgeon now, but right now I'm building a new website + blog for myself, with a simple, typographic layout, 100% static HTML, and as little CSS and JS as I can get away with. Right now a full article (no images) is about 28KB transferred, including a 11KB CSS framework I'll probably cut in half before even minifying it. The assumption now is that >50KB for a page (excluding images) means I've failed.
GP with his 50KB pages would have to get over 20 million views a month in order to use 1TB. Most of us who work in tech can probably afford to upgrade to a larger droplet if that happens.
Hosting your own content isn't free, but it doesn't have to be very expensive, either.
The entire reason that I do it myself though is because my blog is what I use for random technology experiments, so I get some value out of that.
But, let's say that you're distributing a 10MB video with your textual content, and your post does get lots of attention and suddenly you need to transmit 30,000 copies of that video. Let's also say that you decided to use a CDN to host your larger static content, such as videos.
In such a situation, you know how much that 30000 visitors cost you? According to CDN Calculator[0], it will cost you, on the low end, $18 that month, and if you where using Amazon Cloudfront as your CDN, it would cost you $30 that month.
If you're more price sensitive though, you could easily use a Digital Ocean[1] droplet as your sole content source, and in that case it would cost you a whopping $5 that month, as each droplet comes with 1TB of network traffic per month. If that's not enough you could switch to a SoYouStart server[2] for $40 a month, each of which has a monthly bandwidth allowance of 250 Mbps, which is 78.3 TB of bandwidth per month.
So no, you probably don't need to resort to 320x260 video in 16 colors : )
[0] - http://www.cdncalc.com/
Stupid shit... like hosting a video on S3, or really any static website asset on S3.
Let's say I am a photographer and I want to create or share some image galleries. Each image is big, about 10 MB and there may be hundreds of these images (less than 1000). They have a very permissible license, and I want them downloaded and used as wallpapers, for instance. I want the webpage traffic on my site (over flickr) because I want to drum up business and recognition for myself, but I don't want to risk my site going down under hug of death situations. Low cost solutions are ideal because this is a side project.
Is it best to host the main site on a single server (DO droplet) and kick the heavy lifting of image transfer to someone like imgur? What is the cheap and robust solution to this sort of problem?
The web doesn't scale, and that makes it the great centralizer. If publishing ran on peer-to-peer protocols from the beginning popularity would never be an issue.
There is this slight problem about asymmetric bandwidth and ISP-wide NAT, though…
I've been toying around with IPFS lately, but I still haven't figured out how me getting the repo leads to me joining a global network of peers, with data travelling seamlessly between us. I want the latter, just don't know if it's even supported now.
There are some free pinning services (pinning = inform a bot that you would like your content replicated, and then it takes care of it), including at #ipfs on irc.freenode.net, but the network itself doesn't come with such functionality and after I filled an issue on GitHub, it turns out the authors don't think IPFS should do that. Looks to be out of scope for them.
Outside of that, `ipfs add ...` publishes your content's hashes so everybody who knows the hashes can fetch the content from you.
(You also have `ipfs name publish ...` for publishing a website hash to IPNS, which can later be resolved via `ipfs name resolve <peerID>`)
I'd much prefer something like Freenet's automatic replication model. Luckily IPFS allows it, through its own facilities for extensibility -- if we have the will and resources to invest in that functionality.
They do have "flexible" schemes where the bandwidth can increase as demands grow (of course, you pay accordingly), but that's not the default.
(I'm hosted by https://www.gandi.net/)
0.02$/GB which comes out to 20$ per TB and the first TB is included in the price.
As long as you stay away from well-known IaaS platforms, there are plenty of places where you can get bandwidth for less than $10/TB. Lowest I've seen is $1/TB. Having actually used a few hundred TB with them, it's a bit slower but not that bad.