Reasons to not ship a PoA network
This story is written in response to “Ocean on PoA vs. Ethereum Mainnet?”. In this article, I’ll argue for why I think it is a bad choice to ship to a Proof-of-Authority network for a project that wants to go permission-less in the future.
Proof-of-Authority is an algorithm used in blockchains that delivers fast throughput (tx/s), finality and low fees through a consensus mechanism based on identity as stake. Both Geth and Parity can easily configured to run on POA. It allows projects to orchestrate their own networks that run in parallel to the Ethereum main chain. POA chains can be configured to be public or private. POA has the following positives:
Finality (time for a block to be confirmed), throughput (e.g. tx/s), latency and fees can be configured, hence compared to the Ethereum main chain performance is significantly increased.
The Ethereum client running the POA can essentially be customized such that new use cases can be fulfilled. In the case of POA.network’s xDai, Dai is the native currency. Transaction fees are paid in Dai. Very cool.
POA chains benefit Ethereum. Its main chain is not scaling at this point. Even though Plasma (e.g. LeapDAO) and L2 (e.g. Raiden) are shipping, many see sidechains (e.g. POA networks) as a viable scaling mechanism.
I introduce Proof-of-Authority chains and their positives. The next section dives into the problems of POA networks.
Firstly, I’d like to say that there’s nothing inherently wrong with shipping on a POA chain per-se. It’s more about the intend behind shipping to POA chains.
Are you orchestrating a POA chain in your company to issue private tokens for your employees to send around. Great! Are you shipping a public plasma-enabled POA chain that allows users to exit the chain fairly. Cool!
I’d like to argue, however, that when your goal is to arrive at a permission-less, decentralized and transparent network, you shouldn’t ship to POA first. In the following sections, I’ll argue for why that is.
Proof-of-Authority in a public setting is a bad choice, as it has different security guarantees compared to a PoW or PoS network.
Users have to trust in your wet code. Or simplified: How nodes are run is essentially defined in legal documents. Node operators agree to run the nodes according to these documents. If they don’t comply, they’ll face legal issues. To my knowledge, no precedence has been set in court yet. Note: I’m not a lawyer. Building a system with wet code leads, however, to the following properties: (A) It’s not permission-less as in: Everybody can, without the incurrence of a cost, to innovate or influence on any part of the infrastructure.(B) It’s not necessarily transparent as in: Everybody can follow the decision-making processes. (C) It’s not open as in: Everybody can choose to join the network without facing gatekeepers.
Identity at stake is not a strong incentive for businesses running nodes. When the only skin-in-the-game is identity, many node operators fail to even get off the ground. An example: Assume you want to run a POA network that’s about storing data. You get lucky and convince the Internet Archive to run a node on your network. You promise them: “By running a node on our network, you’ll easily break even by cashing-in on the transaction fees”. Now here’s what would happen: The Internet Archive is not in the business of running blockchains. Their core business model doesn’t depend on it. That means, in case they fail to run a node for some reason, they have almost nothing to lose. OK, then say you convince an aspiring blockchain startup to run a node. They’re in the business of making money on blockchains! Well, even though they might run the risk of suffering a reputation-crisis, that won’t affect their bank account much. Depending on their business model and product, they might still be able to make money. Alright, then let’s assume you score one of your customers to run a node. Well in that case, they might actually have vested interest in keeping the network up and running. Generally, I’d say then: A node operator needs to have significant money at stake (or skin in the game) for them to be acting in the interests of the network. Hence, PoS or PoW is usually a good choice in a public setting.
Collusion of entities is likely. To join a PoA network, entities will have to email their public keys back and forth. It makes total sense for them to stay in touch, as they’re responsible for keeping the network up and running. That, however, removes the anonymous nature of running a PoW or PoS network. All of a sudden it becomes very easy for entities to collude. Of course, they’re legally bound not to, but as we’ve seen from the previous points, they might not be incentivized through stake to act in the interest of the network.
Token-bridges are centralized. There’s a reason for Plasma to exist. Essentially, it allows users to exit the chain in case of e.g. a (Byzantine) fault. In Plasma, users don’t need the permission of any authority to exit. Exits are facilitated through an on-chain contract that’s tracking balances. Token-bridges, in contrast, are likely hosted on cloud providers like AWS by the same organization that’s potentially also running a node. This means that when a user transfers money to a POA chain, they have to not only trust the organizations running the nodes but also organization running the token-bridge!
Developers that ship to a POA chain will write different code than developers shipping on a main net. The security guarantees are simply different. On POA, developers can easily take short cuts without spending much time thinking about the horrors of adversarial open networks. Speaking from my own experience, when security guarantees change, the code is usually very costly (time and money) to port.
Now that I discussed the security guarantees of a POA network, let’s move on to UX.
User (and Developer) Experience
Running a POA network is going to give users bad experiences for these reasons:
Let’s be honest. Today’s users struggle to understand what blockchains are. In deploying a separate network (e.g. POA), users will have to additionally learn what it is. They’ll have to manually set the endpoint in Metamask. They’ll have to always remember when to change the Metamask endpoint and for which application. They’ll have to understand what a token-bridge is. They’ll have to do the accounting for how much of their money is on the POA net vs. main net. Some apps won’t allow setting POA network endpoints. They’ll have to understand that even though POA addresses look the same as main net addresses, when sending money to a wrong address all their ETH could be gone. Should I go on? Ethereum user interfaces are built for the main net. User education happens on the main net. In terms of UX, POA networks are the FTP protocols of blockchains.
Smart contract composability. IMO, the Ethereum killer feature. Essentially, the reason for why we have DeFi today. While token-bridges allow for cross-chain value transfer, transfer of state isn’t possible and likely won’t be in the future. I’d like to make this extra clear: A contract deployed on a POA network will not be able to interact with a contract on the main network through state (e.g. calling functions). Yes, value can be sent through a token-bridge and that means one could potentially send ETH/ERC20 from: POA contract -> token-bridge -> e.g. Uniswap. Most DeFi requires the calling of contract’s methods, however. Not having smart contract composability leads to the following: (1) Developers will not be able to plug into your contract APIs unless they also ship on your POA network. (2) Your projects will not profit as much from open-source collaboration, assuming most developers want to ship on the main net.
POA prohibits layer 2 scaling. Most layer 2 scaling solutions are developed specifically to solve the problems of the Ethereum main net. On a POA (and sidechain), you’ll not directly profit when these technologies get rolled out. Instead you’ll have to spend significant time and money to customize them.
- Building tools and deploying a POA network is work. In fact, it’s a distraction. The money you spend on customizing and deploying your POA will be wasted once you ship to main net. As outlined before, it will put the engineers in the wrong mind set. It’s distracting them from shipping the best product they possibly can, I’d argue.
I outline a few possible solutions that remove POA (but not side chains) from the equation.
Ship to main net and wait for e.g. ETH 2.0/Polkadot. Least costly. Best security guarantees. Bad performance. UX will be painful but improving rapidly as you can profit from the community making progress (e.g. meta-transactions). Full focus on shipping product though. No distractions. You’ll profit from composability and projects integrating.
Ship to main net and investigate alternatives to POA that offer better security guarantees to users (e.g. Loom Network is shipping dPoS — node operators actually lose money when they misbehave). Not that costly, I imagine. Better performance than main net. Better security guarantees. UX will be same as POA. You won’t profit from main net innovation as much.
Ship to main net and investigate Plasma as an alternative. Most costly and complex, I imagine. Good performance. Better security guarantees than POA. You’ll profit from the innovation going into Plasma. UX will be as bad as POA but through community innovation will increase at rapid pace.
Ship to main net and consider L2 via e.g. Raiden. Costly as contracts have to be adjusted. Good performance. Better security guarantees than POA. Innovation in UX through community at rapid pace.
All the solutions outlined come as “off the shelf” to some degree and with security guarantees similar to main net. They’re all more or less shipping today.
In this post I argue for why shipping a public POA network is a bad choice when planning to go permission-less in the future by highlighting the security and UX challenges users will face. I give alternative solutions to POA.
(This article was originally posted on medium.com)