Overlay Networks are VMs for the world-computer

4 min readOct 20, 2022

Overlay Networks are virtual machines which allow you to program at the network layer, and create beautiful experiences for developers and users. Here, we are considering unconnected L1s like Ethereum and Polkadot as Underlay Networks, and programmable interoperability protocols like Axelar as Overlay Networks. For an intro to Overlay networks, see this article.

Co-written and co-dreamed by Postman and Jackson Virgo @_nEquals1

Overlays are virtual machines

Just as a Docker container sets up an environment which behaves uniformly regardless of the operating system or hardware it runs on, an Overlay Network lets us define strict behaviors which the raw network itself (the Underlay) doesn’t support. The process running inside a Docker container is sheltered behind invisible walls that stop it from seeing its “true” habitat, allowing it to live in a unique world optimized for its own behavior.

Overlays are dreamworlds which hide the harshness of reality from us and our applications. Source: The Truman Show (1998)

Likewise, an Overlay Network is a map for network traffic which is optimized precisely for that specific kind of traffic — a specialized view of the network. Caches keep popular traffic from having to travel too far; routers direct traffic through the fastest or widest paths, and security overlays, such as Axelar’s rate-limiting feature, can ban malicious traffic before they hit their target.

Keep complexity as virtual as possible

An Overlay Network allows us to program a map of paths that can be taken through the underlay which are optimized for availability, performance, reliability, or security. This map is continuously being updated as network conditions change in real time.

Think of separate networks, like Ethereum or Polkadot as graphs (worth watching if you don’t know what graph means in this context), and an Overlay Network as a graph of graphs. On the Overlay Network, a single node might represent the ability to take a loan, where in reality that node represents the ability to take a loan from many different chains. Since the nodes and edges in an Overlay Network are virtual, you can reprogram what appears on their graph based on user demand, without needing to touch the infrastructure layer.

Once an Overlay Network is built, you can optimize the underlay as infrastructure, rather than both infrastructure and tailored to a specific use case. This is the future Axelar have anticipated, by building interoperability which supports a hub and spoke underlay model from the start. They are leap-frogging a messy stage of point-to-point underlays.

Overlays allow you to virtually get the same connectivity of full point to point, but without the infrastructural fragility and cost.

Check out Honk if you like Hub & Spoke for more of these graphs and how to think about them in the context of cross-chain

Point to point requires a fully connected graph, the most inefficient graph one can build. Conversely, a single hub achieves minimum connected edges, and maximum efficiency at the more costly infrastructure layer.

The real fallacy of the point to point cross chain infrastructure models (most except for Axelar, Cosmos Hub and Celer) is that they try to solve a problem (routing between networks) at the underlay instead of the overlay. They are basically guaranteeing they will be disrupted.

Overlays scale network entropy

Entropy is the maximum amount of information that can be stored in a substrate. A bit is generally the fundamental unit of entropy, 0 or 1. Each bit can store 2 pieces of information, but an overlay encoding can combine the bits together can store exponential amount of information (2 x 2 x 2 x … = 2^n). This exponential scaling can also apply to networks, but you need an overlay layer to be able to take advantage of this.

This theoretical maximum of entropy is limited by stabilisation protocols like error handling schemes, and in the case of overlay networks, these include rate limiting, redundancy of computation resources, finding the best price for a type of service etc.

Communication scaling on the Internet before and after overlay networks.

We keep adding L1s, but we’re not noticing any difference???

Add a server to the Internet, and you don’t necessarily increase the entropy of the Internet. The server needs to be integrated into an Overlay Network to be maximally useful. We’re seeing the same thing happen in crypto now, as the proliferation of L1s doesn’t really improve user experience at all.

Overlay Networks will bring the next scaling payoff

Once you have enough servers and networks, the payoff for connecting them more efficiently vastly outweighs the payoff for adding a new server. Adding servers creates linear improvement, Overlay networks unlock exponential growth.