Why don't more software projects use BitTorrent for downloads?
It seems like it would be great for open source projects with limited funding to use BitTorrent for downloads. However I only see a handful of projects using this model, e.g. LibreOffice and GIMP. Whereas a lot of Linux-based OSes just host their disk images over HTTP, which I imagine could incur substantial hosting costs over time.
Why don't more software projects do this? Are there any big downsides to distributing torrents as opposed to traditional FTP/HTTP downloads?
It's not really worth it for small downloads. I looked at one of my projects and it is an 880kb download.
If you are talking a gigabyte or more there is a big advantage in terms of (1) a completely reliable download, and (2) being able to split up the download into multiple streams. At home I have two ADSL lines and a load balancer and with bittorrent I get double-speed downloads.
GitHub/GitLab releases makes it easy, free, and has virtually no overhead to distribute software to millions of users globally. Setting up torrents and making sure they stay seeded costs time and money, and torrents are less convenient for most users.
Unfortunately web browsers don't support it, so it requires an external client to download.
There's a ton of free HTTP mirroring available for open source; BitTorrent really has no advantage. Bandwidth is so cheap in general that BitTorrent is not needed.
Without Browser Support BT Requires additional client.
>which I imagine could incur substantial hosting costs over time.
And bandwidth is extremely cheap. ( Look at Cloudflare ) Lots of sponsorship and help for Open Source project or if you want to host it on Github for free. Bandwidth is also getting cheaper. May be there is a floor somewhere, but we dont seems to have that in sight.
Many ISPs give customers much better download speeds than upload speeds. I remember personally having problems using bittorrent and seeding too much destroying the network for all using it.
I'm pretty sure OSS gets a free ride on mirrors, because of peering agreements: if traffic between networks is equal, nobody has a case that they're the stronger party on negotiations, so no money changes hands. If there's an asymmetry, theres' an argument that one side has more demand or need for peering and use that as leverage to demand payment. So smaller ISPs that want to balance out a lopsided consumer traffic, will set up mirrors to protect themselves from this.
Since OSS is free to redistribute, a loose federation of mirrors has set up to better match ISPs who want to send more bytes with projects that need to send a lot of bytes.
> Are there any big downsides to distributing torrents as opposed to traditional FTP/HTTP
Analytics is the big reason Mozilla shut down their community FTP mirrors, and bit torrent doesn't really solve that portion.
The best downside that I've heard, at least for doing this automatically, is that not all users have cheap/available upload bandwidth. As long as you're able to opt in as a seeder, it's great.
For the least friction, you could make the download page use Webtorrent, which is a BitTorrent bridge using WebRTC datachannels.
When distributing over torrent, you need at least one seeder with a good connection. You obviously. So you still need to host a server with your package on it, and configure a torrent client to seed it.
I know a lot of HTTP/FTP storage space providers, or I can just put my binary on Github (Github Pages, or Release asset), or host it on a CDN like netlify. I think there is more solutions than real numbers between 0 and 1.
But I don't know a single "torrent seedbox" provider. Setting it up yourself is not complicated, but since it's DIY, this alone explains why it's not more wide-spread.
This is something I've wondered specifically for Musescore 4. They're releasing their Orchestral library as a separate 7.5gb download, but instead of using bit-torrent, they're making you download a download manager. Seems so bizarre.
A lot of Linux based OSes also have BitTorrent downloads.
A lot of ISPs nuke anything that looks like a torrent and send angry messages to their customers.
Bittorrent clients for windows and mac are generally malware infested pits of adware. Wherever i try to be virtuous and use one, i have to clean whatever i installed. I am sure i am not the only one.
Why not build your own install using the bittorrent libraries? The security model is more complicated, for one. If i was an architect and proposed that, i would have to understand the attack vectors and present why they are than the well known solution. And taking control of the software update channel is a massive risk.
Second, i have less control of the user experience. With http, i can pay for the right amount of bandwidth, or time on a cdn. If i implement bittorrent, i still have to have buy capacity, i just have a more complicated model for how much to buy.
Suggested updates can be spread over time - and need to be, for canary purposes. I think Android often pushes an app update over 4 or 5 days, by default? Steady state infra capacity, or even better, low priority which and be interrupted, is cheap.
Given the complexity and business risk (people can't download our software! Our binary got hijacked! Two code paths to test?) And the inexpensive nature of mirrors, and the competence of cdns, there are rarely causes where it would make sense.
Raspberry pi OS has both.
Regarding hosting, I am not exactly sure that it changes anything. You still have to seed the torrent, so the file needs to be hosted somewhere. What is reduced is the bandwidth, but you rarely pay for bandwidth.
I do like the torrent option, but it’s a little more work for the person who downloads.
Because I dislike it so much as a user that I never got into it and never would have thought to propose it.
In Germany, before Fibre (and lol who has that), upload speeds for DSL and Cable have usually been only 10-20% of download speeds, and I know a lot of people who have stayed away from it for that reason alone. Yes, you can limit it, but it kinda kills some of the whole thing. Also the hassle with port forwarding etc.pp. I think I didn't have a client installed on any machine for the last 10 years.
A) If you don't use the cloud, bandwidth is so dirt cheap that nobody cares.
B) Some ISPs will throttle you heavily if you upload too much.
C) Some users still have to pay fees for upload traffic.
There are so many free CDNs and mirrors, especially for hosting open source.
Also most open source projects aren't particularly bandwidth hungry, why bother?
I suspect a big part of the reason is "more work for very little gain" since most users will just download over HTTP given the choice.
I'd love to see some stats from somewhere like Internet Archive that offers both, it'd be great to be proven wrong, but I suspect a number basically 0% after rounding of their data is downloaded via peer-to-peer BitTorrent traffic.
A few companies I've worked at have used torrents to roll out new service binaries (multi-gb tarballs of Java jars) and periodic index deployments using torrents.
The biggest advantage is just diffusing a thundering heard without increasing latency considerably.
Shout-out to tTorrent.
Maybe not for individual software projects, but I would like to see something like this for installing and updating packages via the OS's package manager. Given the fact that package repositories have mirror servers, it should definitely be possible.
Blocked or otherwise impeded in some places: eg, unable to request peers from a tracker.
Spotify used to use BitTorrent, or something closely related.
Back in the early days, colleagues in my office would be falsely accused by IT of file sharing, when the only crime they were really committing was listening to Jack Johnson.
Until everyone has synchronous internet it won’t need to be a thing. I’ve pointed out many times that between BitTorrent, tor, and bitcoin we’ve already done alleged web3, which is what we used to call the Dark Web.
Windows 10 claims it distributes updates in a peer-to-peer manner. Or at least it used to. Does that still happen?
How is it better than HTTPS+CDNs both performance wise and implementation/maintainance complexity wise?
Valve hired Bram Cohen back in 2004, and AFAIK it is/was core to Steam's distribution system.