Hi, author here. Thanks for checking out Bifrost. The short pitch is: any protocol over any transport with seamless cross-platform (browser <-> native) communication.
Building internet applications often involves writing code that is tightly integrated with the choice of communications transport, for example, one has to make the design decision to use either WebSockets or WebRTC, and then all other design decisions are driven by that.
Bifrost aims to allow developing applications without knowing or caring which protocols are used under the hood. The app can just say "Open a stream with peer X" and whatever configured transport(s) make that happen.
I'm still working to improve the docs and new user experience, if you have any suggestions on how I can make it easier to understand, I'd greatly appreciate it!
I see, thanks. I think it would be useful in the docs to describe both some less technical as well as some technical examples of useful applications of your work.
Do you have some examples of _concrete problems_ that would be easily solved with this? If so, mention them.
Telling folks about technical aspects doesn’t help advertise your work unless you can help a new reader relate it with their own experiences or problems.
Bluetooth could work! Just need to implement the Transport interface.
Peer discovery is not implemented at the moment but could easily be added (and will be soon). Here's how: any controller can resolve EstablishLinkWithPeer. A peer discovery controller could compare the list of discovered peers with the list of desired connections and initiate connections via UDP when there's a match.
Signaling is managed by the Link implementation. For UDP we use quic-go.
This stack could be the basis of a software to share things no matter your connectivity situation.
You fire it up, it scans things around. Local network? It uses that. Not local but internet, it uses that. Nothing but BT, you can still detect people around, chat and exchange files.
People could be identified by public keys and that's it, not need for an account, signaling done by zeroconf/bt-scanning/bittorent.
Basically a solution to the current chat nightmare.
But the amount of work is overwhelming, even with this.
There is a design for routing in the repo (and tor-like circuits). However it is not implemented yet. I'd love to grow the community and have more people collaborating on this.
Yes, that is not very clear. It seems to me (corrections welcome) that it could sorta work as a application oriented VPN where a VPN cannot be installed.
Something like:
local_app_1--->Bifrost Client1-->Bifrost Daemon0-->OS0-->//network//->> same in reversed order
local_app_2--->Bifrost Client2-->Bifrost Daemon0-->OS0-->//network//->> same in reversed order
instead of:
local_app_1--->OS0-->VPN0-->//network//->> same in reversed order
local_app_2--->OS0-->VPN0-->//network//->> same in reversed order
That is, each application has its own client tunneling to a common daemon which then tunnels using the usual network stack to a similar arrangement on the remote machine.
It's a peer-to-peer "engine" with switchable components. Seems to run on different platforms (browsers, mobile, desktop, server). Basically the library helps two different computers to talk with each other, without any servers setup in the middle.
At a glance, it looks pretty much like libp2p (https://libp2p.io/) but seems to integrate with libp2p as well (meaning you should be able to use Bifrost on one end, and libp2p on the other), so I'm guessing there is at least some fundamental difference, but I cannot spot it. Seems to use slightly different terminology compared to libp2p.
I was a little confused by this...maybe i'm not expert enough at networking. I initially thought this Bifrost was related to the matrix bifrost (https://github.com/matrix-org/matrix-bifrost) ...but don't think that is the case...I think because *this* bifrost seems lower in the network stack, whereas the matrix bifrost is higher in the stack, and more dedicated for puppeting users via bridging between different messaging networks (moreso at application layer)...?
Dynamic configuration of components. It uses a system of communicating controllers which is more flexible in terms of runtime reconfiguration. libp2p is an excellent project and can be integrated into Bifrost's system as a library.
Check out the repo on GitHub as well as the others on the same GitHub organization! Also feel free to join discord and ask questions or shoot me an email. https://aperture.vision
The README more or less describes the "what" but there's no "why".
Why is this needed, who is this for (target audience), why did you build it, which gap is it closing, which problems does it solve in a homelab, or startup, or enterprise, i.e. "what's in it for me", how is it different than alternatives?
I feel like git should have a standardized howto.md or something that gives you some version of instructions on how to use the software in this repo. Especially in a case like this where there's no context as to what this thing needs to be used for how it's useful why I should use it what the advantages might be etc.
I'm not a web developer. None of this is intuitive or easy to grasp for me. Even the inclusion of the demo and example directories weren't easy for me to find because they aren't standardized. The fsf also thinks that free documentation is important. https://www.gnu.org/philosophy/free-doc.html . I view software like this as pretty exclusionary. We will have to agree to disagree, but I would say that the amount of people with requisite experience to understand software is lower than the amount of people who are (or in my particular case, were) interested in learning about it by looking through the source. This interaction is a prime example of why the free software movement stalls. Software written by insiders, for insiders is exclusionary from my point of view.
> Software written by insiders, for insiders is exclusionary from my point of view.
Where do you draw the line? At one point, you have to accept that some projects require deep and context-specific knowledge. I can't be expected to be handheld if I dive deep into some math libraries without knowing the prerequisites. And if I do anyway, it's up to me to figure out at least the basics on my own.
The README has the examples and explanations you're looking for, they're just using the language common in the peer-to-peer ecosystem, as it's a peer-to-peer library/cli/daemon.
...which is kind of what is being complained about by GP. I'm not going to pretend that anyone who can actually read this readme has enough expertise in p2p networking to remake bifrost (obviously it's quite a thing, if it does what it says it does) but at a glance it certainly feels that way.
You write the UDP payload onto PCI-Express, then the NIC writes it onto the wire as an Ethernet frame.
Obviously you're limited by PCI-Express bandwidth then Ethernet bandwidth. There is also a bit of jitter tied to the underlying protocols but those delays aren't really significant for most applications.
Meanwhile, TCP can queue data indefinitely, based on how the other end is sending acks and the size of the window.
And what about all the switches, routers, bridges, firewalls, and whatever else in between endpoints?
What a strange thing to require. It comes off a bit like X-Y. You seek high performance. But you state it as wanting zero copy networking. There’s a lot more to performance across the overall system than that.
