At Chorus One, we have a strong conviction in the potential of a multichain future. We believe that specialized blockchains play a crucial role in discovering and nurturing new use cases, and ultimately driving mainstream adoption. Since joining Chorus One about two years ago, I've been pushing for us to do the same in the Avalanche ecosystem as these two ecosystems have similar visions of what the multichain future can, should, and will look like.
Last year, we entered the Avalanche ecosystem. Our work will only intensify in the coming years with Chorus Ventures, our ventures arm, investing in native Avalanche projects. We also use our expertise in tokenomics and infrastructure to help projects launch their permissionless subnets. We will be presenting this topic both at the online Subnet Summit mid April and the Avalanche Summit II beginning of May.
In my view gaming is key to onboarding the next wave of users and a fundamental step in the road to mass adoption. This article aims to present the exciting future of blockchain gaming and demonstrate how the Avalanche architecture, particularly the multichain subnet architecture, is the ideal substrate for this vision. Through a two-part series, I will illustrate how one can develop an unstoppable game. By unstoppable, I mean a game that not even its creators can censor or stop if one day they move on to other projects. A game in control of its users.
So let's get to it.
To make this exercise as clear as possible I will look at a game I have plenty of experience with having played around 2000h. The game in question is Path of Exile, in my personal opinion the Diablo killer. This game needs no introductions but:
A short video:
Take a look here for some more gameplay videos.
I cannot emphasize enough how deep the economics goes in this game. The reason being that its economics are fundamentally tied to the crafting system for equipment and to the simple fact that you need to craft gear yourself or buy it from someone if you want to reach the endgame. Purely random drops cannot take you there.
Every season a "league patch" is released with new contents and the economy is reset. Characters and loot from previous seasons are still available to play in the "standard" league and the standard league economics are interesting in their own right but as a driver for innovation and to give new players the ability to compete in a more level playing field, these resets are very important.
The goal is thus to envision a version of PoE that is unstoppable and in the hands of gamers. You might ask: why would developers make such a game? To which I answer: the first one to do this becomes a first mover in a technology that soon will be expected from all games. And why will this be expected? Why do gamers want this? Well, this is a game you can continue playing and you can really own it. Like how it was in the dawn of console gaming. Be it real or game money, you can trade assets and no one can censor you. If you recall the anecdote, this was the reason Vitalik started his work in crypto.
In the centralized world, an ARPG like Path of Exile consists of a client/server platform where the server infrastructure is run by the game developer, and where the client freely available or purchased in a marketplace. Next, we will look at the features and responsibilities of the server side as this will be where our decentralization efforts will mostly focus.
The ability of client-side tampering with binary can cause all types of attacks/cheating. This is an arms race but currently, it is tackled via lock-step state validation. More on this later.
Most games need randomness. For anti-cheat reasons, this is taken care of at the server level. In the case of PoE (and ARPG more broadly) this is even more important as loot, damage, map layout, and even AI are parameterized by random inputs.
Of fundamental importance for a healthy in-game economy is that the more powerful items are, the rarer they should be. That is, their drop rate should be lower. This is accomplished by drop-rate lookup tables that are set and maintained by the server. Again due to anti-cheat measures, it is the server that, when appropriate, generates a random drop.
Even in PoE which tends to be dominated by PvE (player versus environment), there are situations where players interact: regions where PvP (player versus player) are allowed and sanctuary environments also called player hubs. These interactions need to be facilitated by the game server.
Special trading windows and functionalities are implemented so that players can exchange goods in a safe way.
Looking at the above set of functionalities that the server must provide, we can identify three different types of backends that the server infrastructure needs to maintain. These are the components we will need to "permissionlessly" decentralize. The following figure gives an idea of how the server-side interacts with these backends and the client (overlap indicates communication).
There is a need for queryable databases, with loot tables clearly being one such need. But many more are present: leaderboards, player info, skill table, effect mechanics, and many more.
A key-value store that can deliver monolithic "chunks of bytes" is also a necessary backend. The game needs to ship itself and its updates with a big proportion being graphical assets. For this dedicated content-delivery networks are employed.
As mentioned before the server infrastructure needs to be able to keep clients in sync across PvE and PvP both for anti-cheat purposes and for facilitation of user interactions.
So why is Avalanche especially suited for this exercise? How will the architecture of such a game change and what technologies do we need to leverage to accomplish our goal of a decentralized, unstoppable ARPG game?
Avalanche has two genius breakthroughs in its design: its consensus being the first and the subnet architecture being the second. The latter is highly dependent on the former. Let's see why.
Avalanche consensus is without a doubt the most advanced consensus out there and is correctly categorized as a third type of byzantine fault tolerant consensus following the discovery of signature accrual and Nakamoto consensus. It is the first meta-stable type of consensus algorithm. This consensus enjoys enviable properties: it scales easily with the number of validators, it is leaderless, and single-slot final. I won't go into much detail but suffices to say it accomplishes all of these by being a consensus algorithm based on a statement about an emergent property of the system. Let me explain what I mean. You can think of the network as having the property of being consistent (all validators agree on the current state). In Avalanche this property is emergent. Like the temperature of a gas, it exists as a property derived by the local interactions of its constituents “particles”. In the case of the gas, particles bouncing of each other exchanging kinetic energy in their small neighborhood gives rise to the macroscopic property of temperature. In Avalanche, validators are the particles and contrary to other consensus mechanisms they interact only “locally”, that is, with a small number of validators that are randomly selected in each round. Somehow - and here there is a strong mathematical theorem behind it - this is enough for the network to have a well-defined sense of state history. Even in the presence of attackers.
It is the property of essentially limitless scaling in the number of validators that allows for the second genius move. You see, Cosmos is the originator of the concept of an app-chain. In this design, it is absolutely necessary that chains can "talk" to each other to really cover all the use cases one is interested in. For this reason, they developed the IBC framework. This is an elegant framework for trustless communication but it incurs a significant requirement to a prospective chain: as a destination chain you need to keep consensus information of any given source chain you want to communicate with in the form of a light client. Wouldn't it be ideal if this information would be globally available to all chains from all chains? This is impossible with a limited set of validators.
So to have an unlimited set of app chains that can trustlessly communicate without having to keep light clients of every other chain they communicate with you need an unbounded set of validators in a global chain that keeps all this information. I hope you see where this is going: this is exactly the subnet design.
In Avalanche the main network that every validator must secure contains three chains. The P-chain (Platform chain), the X-chain (eXchange-chain) and the C-chain (Contract chain). The X-chain - which us currently a DAG but will become a linear chain in the near future - is a chain made for throughput exchanges of assets much like a blazing fast Bitcoin network. The C-chain is what most users are more familiar with and is an EVM based chain. It works just like Ethereum but faster and with instant finality. Great. But the real genius comes from the P-chain. This chain tracks all validation related transactions of the mainnet and all subnets. This is what will enable the unbounded, composable network of app chains. Since all validators have the P-chain at hand, any two subnets can communicate directly provided they want to. In IBC, on the other hand, with its hub-and-spoke design you have the unaddressed issue of path dependence.[^1]
So, we will leverage an Avalanche subnet for our game. Main reasons are the excellent scaling properties of its consensus and the application-specific, isolated nature of the subnet approach. On top of that it supports cross-subnet transactions allowing for valuable assets to move around freely in the ecosystem. Finally but not least, there is also VM2VM message passing that allows the validators in a subnet to easily check the state in other connected VMs be that within the same subnet, in the mainnet, or another subnet running in the same validator (the latter has not even been explored yet).
An Avalanche subnet is essentially the following:
The set of validators is dynamic but can be either permissionless or permissioned. The specification of a blockchain is comprised of a specification of a subnet this blockchain pertains to and the specification of a VM (i.e. virtual machine) that characterizes the valid state transitions in that blockchain.
Ava Labs recently announced HyperSDK a toolkit not much unlinke the CosmosSDK to help developers easily build VMs to power their subnet. From now on, they can focus on the logic of the application and worry much less about synchronization, consensus, state storage and availability and other blockchain-heavy topics. On the other hand, if you want to, you can customize these aspects as the SDK was build with modularity in mind.
As mentioned before, our intent is to decentralize the game. For this, we will need to decentralize the server infrastructure, mainly the three points named above: databases, content delivery, and anti-cheat logic. This will be done by defining specialized VMs and the corresponding blockchains for each of those game infrastructures. All of this is packaged in the game server binaries which will be run by the validators in the subnet.
A game client will essentially be submitting transactions to the server network. Clearly, the game client is responsible for client-side rendering which is something we do not need to bother with on the server side. In terms of execution hardware, the game server is much lighter than the client and we will exploit this.
Keep in mind that being a player does not mean you can’t be a validator as well or a delegator to a Chorus One validator ;). This is obvious but worth mentioning as this means that for the first time ever a game can actually be in the hands of the players. With governance, even the game features and roadmap can be decided, paid for, and rolled out completely in a decentralized fashion.
So the big question: what are the blockchains, VMs and technologies used for this purpose? We dive in.
The BlobVM already exists in an advanced prototype stage. It was developed by Ava Labs and is available in open-source. What it does is provide a dedicated, seamlessly integrated (at the subnet level) content addressable storage with customizable parameters regarding permissions for read/write and persistence.
We use BlobVM for storing all art, texture, and models, i.e., all game assets. Even the game client binary can be updated via this method. In a fresh install, an externally downloaded game client connects, and sends a transaction to download all necessary assets. Note that this transaction could be a way to monetize the game but this is optional of course. In other words, this transaction would give you a game license NFT.
Now as mentioned before we want to give power and value to the gamers. Path of Exile is famous for its rich economy and is a formidable laboratory for NFT tokenomics. By giving the gamers the option to mint any found loot item we give this economy real value. There is plenty of opportunity and pitfalls here to fill in another article but it is important to mention that PoE works by having multiple “leagues” which give an opportunity to always “reset” the economy and give chance for new players to “make it”. We think this is an important aspect to keep in the decentralized version of this game. As an example of how we could explore this, we can configure it so that minted NFT only work on the current and previous leagues.
For tracking a gamer’s collection we use AVM, the Avalanche VM, which is a DAG (directed acyclic graph) based on the UTXO model capable of massive throughput. In fact, this is the underlying VM of the mainet’s X-chain. Note that since the announcement of Cortina (the next dot release of the Avalanche validator client) the X-chain will move from being a DAG into a linear chain. Here we have the option of launching out own AVM chain for assets transfer or, use the X-chain directly which would make all of the game’s NFT directly available to the wider Avalanche community (NFT reuse in games is an under-explored area). The AVM supports ANTs or Avalanche Native Tokens that can easily be imported/exported across the majority of supported VMs as it defines a unified API for cross-chain atomic swaps.
PoE is a free-to-play game that monetizes itself via cosmetic-only user-purchasable content. This can be easily supported via the AVM chain as well. Simply: an NFT in the user wallet ”unlocks” these assets to be delivered via the content delivery mechanism. This is essentially a VM2VM communication as is desirable and quite probable that the X-chain will support account lookups via this mechanism.
As mentioned before, as with any modern application, the game needs to store global relational data. For example, loot tables, league-specific information, game metrics, user metrics, NFT market data, etc. The list goes on and on. For this specific use case currently, many web3 projects use The Graph: a sophisticated but complex decentralized solution. A few issues arise with this approach:
Because of these, we propose a new type of VM we dub SQLVM and this will be the topic of our next article. But in a nutshell, you should think of it as a hybrid between a app-specific indexer and a persistent relational data store.
It allows for specific types of transactions that query/write to a globally replicated ACID relational database. Here we automatically benefit from the fact that blockchain transactions are atomic at the consensus level which makes designing the underlying database much simpler. For example, a suitable design can be done for a VM where the runtime state is an instance of any query engine: row-oriented like Postgres, column-oriented like BigTable, or document-based like MongoDB. Keep in mind that even this is overkill as we don't need their replication features. What we need is their query engine and storage solution. Most of these databases have sophisticated query planners than can take the place of fee estimators. The beauty here is that Avalanche will take care of maintaining this database eventually consistent which suffices for our use case. More sophisticated designs are certainly possible. The job of the VM here is to essentially declare the types of transactions (write/reads), the fees and verify the blocks by applying the transactions in the database and updating certain database hashes (will be needed for anti-cheat below). For our game - or any other app chain using this backend - other VMs in the subnet should be able to read the database at will which can easily be done with VM2VM.
Similar to how a non-validating Avalanche node have access to the mainnet state, a game client could be a node of this chain running in non-validation mode so as to keep this database state at all times for easy synchronicity.
Now to the technologically most innovative piece of the puzzle: to run anti-cheat as a ZK verifier. This is such a breakthrough technology that it would be an improvement over existing anti-cheat technology on centralized games.
Anti-cheat works, as mentioned before, as lock-step game simulation. What this means is that the game client is essentially an input system and a rendering engine of a game that is actually run remotely on servers. This Introduces latency which is the reason why game server farms have to be deployed across internet “regions”. ZK changes the game as it allows one to codify all game state transitions in a prover which we can run on the game client (remember the gamers tend to play with machines that are quite powerful) while the server is just a verifier! This has the added benefit that it even liberates the server from having to run in lock-step, to begin with! Essentially we can use eventual consistencyto catch the cheaters. Put differently, we don’t care to verify every little state transition that happened but batches (or recursions) encoding all transitions that happened in a configurable time window: 1 second, 10 seconds, a minute, an hour…
It is obvious what a powerful idea this is: no need to simulate full-blown games on the server. For example, we now can use more sophisticated AIs in the client. The fact that you have to run the game on the server is one of the reasons no modern AI is in use in games. Why not use GPT-4 for creating procedural quests??
We will have more to say about a ZKVM in a future article but I would like to state a few things. Firstly, note that we are not even using the zero-knowledge aspect of this VM and this gives more freedom in the exact construction of the protocol. In precise terms we are interested in SNARKS not necessarily ZK-SNARKS. Nonetheless, we expect that applications that use this zero-knowledge aspect will also exist.
Secondly, we might not be at the stage yet where fast enough provers exist to prove the state transition for a game like PoE. I'm not an expert, but I expect that schemes leveraging the GPUs in the gamer's clients will be just a matter of time.
And finally, we are talking about a very specific VM - that of the game - and not a generic programmable one like the EVM. We need a prover for those exact transitions that happen in game. This is potentially another route for optimization.
We hope to have convinced you that the future of decentralized gaming and player-owned gaming is bright. When Vitalik joined the crypto movement I don't think he thought his dream would come true on another chain, but I think he will be satisfied nonetheless.
But more importantly, we hope the reader is also convinced that this is only possible in a clean, elegant, and reusable way via the subnet architecture. Sophisticated applications like this will only flourish when good reusable VMs are available much like reusable contracts are right now. Multiple VMs demands multiple chains in a subnet architecture. Although technically possible to cram all of these backends into a single block to be serialized/deserialized and verified using a single chain, this would not only hurt code reuse but is also impractical since it is clear that these backends might need different blocktimes.
Of course, there are a lot of unknowns to this as I am not a game developer. I just want this to jump-start the imagination of developers in general (not only game developers) to the reality that the future is app-specific multi-chain subnets. And so that someone develops an unstoppable ARPG like Path of Exile!!
Tune in for some follow-up articles on where we attempt to detail somewhat the SQLVM and ZKVM and come talk to us in the summit. See you there!
Chorus One is one of the biggest institutional staking providers globally, running infrastructure and validating over 40 blockchain networks. Since 2018, we have been at the forefront of the PoS industry and now offer enterprise-grade staking solutions, industry-leading research, and also invest in some of the most cutting-edge projects through Chorus Ventures. We also invest in subnets on Avalanche so if you’re building something interesting, reach out to us at firstname.lastname@example.org.