Cracking down on that mesh network hype 😤

Decentralized, self-healing, prevents censorship, reliable off-grid, decentralized… did I mention decentralized? OK, well, decentralized.

Mesh networks have existed for a long while now, but recently they’ve seen a re-emergence of popularity through various home-use Internet routing products and the rise of mesh networking-based messaging apps used in the Hong Kong protests.

But why exactly should we care about mesh networks when today’s Internet provider services work just fine for us? Why does this whole decentralized network thing even matter?

Most importantly, how does it work?

[confused_pikachu.jpg]

The Status Quo

Today, the most used network topology is the star topology – wherein endpoint devices (so things like laptops, phones, printers, etc.) will all tap into one central base station. This could mean a cellular tower, a DSP Internet modem in individual households, or any wireless access point throughout any metropolitan city.

You can think of this in the context of a female dog with 6 newly born pups. A star network topology would mean that all the pups need to feed directly from their mum, or else they won’t get any milk.

A mesh topology, on the other hand, means the work is instead distributed across the entire network. All you need is one node (typically a router) that’s physically wired into a network connection, and then all the other nodes will wirelessly bump the data packets along to their destined receivers, like a glorious interlocked spiderweb of data. In this way, each node is connected to every other node.

Back to the momma doggo analogy: this means that the pups will create a massive industrial milk supply chain, pulling milk from mom in increments in a concerted effort to pass them along in containers to the rest of the family.

Meshy-meshy

There’s a couple of key components that make these networks different:

  1. Self-healing properties: Even if one router node drops out, the network will just reconfigure itself automatically because every node is connected to every other node (at least in complete mesh networks). This is called fault tolerance.
  2. Inexpensive pricing: Buying a bunch of routers costs a lot less than building a single base station. Pretty self-explanatory!
  3. “Last mile” connectivity: This is the problem where endpoint devices at the edge of centralized networks often have a way slower connection due to the bandwidth overload: there’s too many endpoints and too little star “centers.” This can be solved a couple of ways, mainly by installing a crap ton of wires, which is a lot pricier than simply deploying a mesh network to pass data on to that last leg of devices.
  4. Routing protocols: This is mesh networking’s kryptonite. The routing protocol indicates a set of parameters by which the network finds the best possible route to forward the data to its final destination. Therefore, the quality of connection in a mesh network depends almost entirely on the quality of the routing protocol.

Point 4 is really the catch: in a world where services like Amazon or Google can lose millions of dollars directly due to an extra couple seconds of latency in response, businesses can’t afford to take the gamble on a network topology that doesn’t guarantee quick responses, especially given how robust today’s network infrastructures are in metropolitan areas.

So you have all these great things about mesh networks, mainly that they can prevent censorship and monitoring due to their distributed nature, they’re highly adaptive to sudden breaks in the network, and are super reliable… But at the end of the day, they’re not exactly commercially viable – especially not now, when so much centralized infrastructure is already in place.

In most Wireless Mesh Networking (WMN) research, the general idea of what a best-case scenario mesh topology looks like is actually a combination of both methods I just mentioned.

You have one (or multiple) network gateways that inject an opening to the outside network, multiple access points (APs) that are robust and don’t have power constraints (these will be the nodes that pass on packets to the rest of the network), and finally a bunch of user endpoints that connect to these APs in a star topology-like formation. This way, you have more of a tree topology than a true mesh topology, but commercial demands for speed can still be met because now you’re not relying on the unpredictability of mobile nodes.

The Future of Mesh Routing Protocols

I don’t think the commonly accepted “tree” topology hybrid makes a lot of sense – especially when it comes to the main non-trivial use cases of mesh topologies, like post-emergency mass-communication networks, extreme rural connectivity, or transmission-intensive industrial applications.

You’re not going to build a bunch of access points after a tornado just hit the city, nor does a slum in India have the resources to do so in the first place. That literally doesn’t make any sense. On the other hand, functionality of the network will still matter just as much to them.

Obtaining the best routing protocol, even when taking into account a network that is constantly transforming and transmuting at every moment – is really just an optimization problem. This isn’t really news – but the way we tackle this optimization problem makes all the difference.

In “Routing protocols in wireless mesh networks: challenges and design considerations”, a 2006 paper written by Sonia Waharte, a WMN is modeled mathematically as a graph G(V, E), where V is the set of vertices (nodes), and E is the set of edges connecting each node to every other node.

At the moment, most conventional machine learning techniques, by abstracting a network’s graph into a flatter set of datapoints, would be losing a lot of critical information about the relationships between each node. But geometric deep learning, a new cutting-edge field in the deep learning space, can potentially solve this problem, as it performs computations on non-Euclidean data – in other words, data like graphs, where the information resides precisely in the relationships between each vertex.


In the advent of 5G, where shorter-travelling signals become the key logistical barrier to adoption, mesh networks may become a key partner to creating the next generation of network connectivity.

Lastly – I cannot think of anything that could be more important than access to information in the 21st century – it’s what allows us to unlock more human potential than ever before. But that access is severely hindered by the network infrastructures that we have today. If we can get all 7 billion of us in the world online – what kind of new solutions to the hardest problems would be dreamed up? And who would be the ones to usher in a new era of prosperity for humanity?

Leave a Reply

Your email address will not be published. Required fields are marked *