What advantages does your project have over Yggdrasil?

Previously on HN (sadly no discussion though)

https://news.ycombinator.com/item?id=40828774

AFAIK no MAC addresses are involved. It assigns you a new IP address, internal to the network, a bit like a VPN.

some of us even still have Yggdrasil installation media

[dead]

I saw that with Reticulum, Yggdrasil is actually a "Backbone interface" [0] used to connect Reticulum instances. What makes a lot of sense.

- [0] https://reticulum.network/manual/interfaces.html#listeners

I'm stills waiting for the Meshtastic vs MeshCore vs Reticulum thing to shake out.... They all have pretty big disadvantages, most of which seem very fixable.

I think what they really need is cheap solar nodes that have external WiFi antennas, so they can mesh over point to point directional links.

LoRa meshing seems like it's eventually going to be a dead end in any kind of dense environments, WiFi just gets you so much more bandwidth, and the edge nodes only need ~2mA of power.

AFAIK only Reticulum and Meshtastic have software support for that ATM.

I have a proof of concept almost working using OpenDHT as a global routing backend for Meshtastic, but Meshtastic still has some security issues (No authentication on a lot of important stuff), and global routing is really only currently easy to do for unicast traffic.

But if they got rid of the channel hash thing, and instead they gave each channel a special receive address(like IPv6 multicast groups) with their own keypair, global routing would probably work fairly seamlessly.

I think ideally they'd change those node/channel IDs to rolling codes so you can't traffic analyze the global routing.

With MeshCore, it doesn't seem like there's documentation on what info the packets actually give you, so I'm not sure how that would work.

And then with Reticulum, there's no C++ implementation of LXML yet, no standalone devices, or anything like that, so I'm not sure if it can replace Meshtastic/MeshCore until that exists.

Meshtastic is also lacking some applications layer stuff, they have the same issue most IoT protocols have, tons of predefined packets for specific sensors, but no modbus-esque generic telemetry packets, no way to just say "Register 59 is 37.6, look up what that means in the register layout file with UUID Jfjeiuehbdhd"

I think Reticulum is closer to having that, with the "fields" thing.

I'm not sure how the global routing works on reticulum though, I get the impression every node needs to know about every other node with some kind of gossip protocol, but I could be completely wrong.

Reticulum also used Kivy for their mobile app, which doesn't feel as polished as Meshtastic, and it's licensed under some non-commercial creative commons thing that causes issues including it in distro repos and such, or so I've heard.

Meshtastic doesn't exist as a convenient Arduino library or Pip installable package or anything like that, it's not exactly very elegantly embeddable in existing apps, and even if it was, it doesn't have the generic abstractions to actually do custom stuff in a way the app understands.

It almost seems like we either need a heavily redesigned Meshtastic 3.0, or something completely new.

They're all amazing projects, but they're all missing something.

There seems to be a misunderstanding. Yggdrasil is not a single network, it's a tool to make such networks. So if you have some computers with some ability to connect them, you can set them to peer in Yggdrasil, and instantly get an auto-configured irtual IPv6 network. (You can even slap some IPv4-to-IPv6 NAT over it, but we shouldn't mention crimes against humanity.)

It seems to me that the main goal of the project is automatic routing that promises that in any peer graph any peer can reach any other at least somehow, which is fine for many use cases. Only in case of moving big data or very different cost of links you might want to check the node hierarchy and set manual peering from X to Y and Y to Z to make traffic flow the way you want.

Public network is a volunteer-run test ground for real life usage, and a way to connect systems that can't connect directly because of NAT. Internal speed test that is 8 hops away on the other side of the world gives me 30 Mbps, closer one, in 5 hops, reaches 90 Mbps (and the local interface is 100 Mbps), so it has some throughput for casual users.

As far as I can understand, current latency-based routing selection is just the most simple way chosen to optimise the network across the whole world, and prevent random 200 ms hops back and forth to intermediate nodes. If traffic from nodes in one half of the tree not having direct connections to the other half goes through the central nodes, it seems that at the moment it's enough for owners of central nodes to peer with each other, and have enough bandwidth. In the future they might need to take the proportions of available channels into account.

> The challenge with these schemes is always making them robust against intentional attacks.

Right. You join the network by generating a random IPv6 address and telling someone about it. So anyone can generate vast numbers of dummy node addresses, which everybody has to track. Now you need spam filtering.

Another related issue is that decentralized networks are often targeted by attackers more than centralized ones, since a common use case for these networks is sharing information that the central entity (or some power that can coerce the central entity) does not want shared, e.g. organizing in authoritarian regimes, illegal file sharing

Is there no way to blacklist nodes?

I don't know how ygg handles nodes but there's no reason a mesh network has to just accept random new peers, you could do something like mTLS where peers have to present certificates distributed out of band.

Could you solve this by running something like I2P on top of the mesh network? I2P has implemented strong protections against typical attacks.

at least in the original implementation of this protocol (cjdns) there was already a tiny bit of proof of work happening in address generation. forgive me as this knowledge is over a decade old but it involved finding a curvecp private key that scalmults into a public key that (maybe whose hash?) begins with some specific number. this number is what's actually used as the ipv6 address. and that specific number is the routable ipv6 prefix.

the process would be run again and again during configuration generation until a key that fit this criteria was found. one could up the difficulty of this process considerably.. though not in a protocol backwards-compatible way.

you also needed to find a peer.

but yeah that's a gnarly hole.

Ygg doesn't operate strictly as a shortest path protocol. It prefers lowest latency links already and they're roadmapped to handle link cost / dynamic weighting / link congestion. They're targeting a blended weighted metric overlay which would certainly prefer the chain of quick/fat links.

Yea, the website was conspicuously opaque about the protocol. Having lived in the networking and IP routing world for decades, that makes me suspicious that this thing is yet another non scalable system. But hey, I’m willing to be convinced.

Yggdrasil is from Norse mythology. So I’m not surprised that a bunch of different people name software after it.

https://en.wikipedia.org/wiki/Yggdrasil

Just like many might pick cool and important names from Greek mythology for example, like Poseidon.

https://en.wikipedia.org/wiki/Poseidon

https://minecraft.fandom.com/wiki/Yggdrasil

If you care about performance, I don't think any amount or style of mesh networking will outperform traditional backbones.

This reeks of LLM

Yes. The answer is yes, we still need a backbone.