The Modular Blockchain Paradigm — How Blockchains Scale for Mass Adoption
Written by: Cameron Sepahi
On a first-principles basis, first-generation blockchains are distributed ledgers with shared-state that is cryptographically secured. Distributed ledger means a “database” that manifests itself in multiple copies stored by nodes (i.e. computers), shared state means that the state updates that add records to this ledger are validated and executed by these nodes, and cryptographically secure means that this state is encrypted, effectively rendering it impossible to revert or manipulate (i.e. it’s immutable) [1].
First-generation Blockchains
There is a good reason why I use the term first-generation blockchains. Up until today, blockchains performed everything: data availability, consensus, and execution. Thus, we came to know the blockchain trilemma: how can you create a blockchain that is highly scalable (low latency, high bandwidth, and high throughput), without sacrificing security and decentralization (the former being censorship-resistance, and the latter being a participant’s ability to run a node) [2]. Though the crypto industry has adopted many different definitions of decentralization, it is more broadly accepted that decentralized blockchains give users the ability to run a validating node in a reasonably cheap way such that they can independently verify the state of it.
Many chains today claim that they have a credible solution to this decentralization problem, but the reality is that they sacrifice decentralization for greater scalability. These tradeoffs are fine to make — in fact, many users prefer to use these chains because it’s much cheaper. However, the chain that can offer the highest degree of decentralization, security, and scalability will serve as the canonical credibly neutral base layer because of the long-term vision of the industry.
All blockchains offer users the ability to run a validating node, and there happens to be a lot of misinterpretation as to what this means. People tend to conflate validating nodes and mining or block producing nodes as the same thing, when in reality they are different. Often, block producers (miners or PoS validators) run validating nodes to verify the state of the chain, but regular users may run a validating node without being a block producer. These users don’t get any reward for doing so, but nonetheless they’re able to honestly check the chain without trusting external sources.
Users may run full nodes or light clients. Full nodes maintain the entire history of the chain and check the validity of every transaction. Light clients only download the block header (containing the block’s metadata), thus assuming the block contains valid transactions. Though light clients are cheaper to run, they are less secure than full nodes because of the inherent trust assumptions, namely their reliance on full nodes to honestly validate transactions. This is the crux of the data availability problem — how can you offer users the ability to cheaply run a node without introducing centralization vectors from light clients? [3]
Current Tradeoffs
Let’s unbundle the reality of the crypto industry — no blockchain has made a 10x improvement in scalability — Yes, not Ethereum, not Binance Smart Chain, or Solana, or Avalanche, Luna, Fantom, [insert your favorite chain here]. If they claim they do, they are either being dishonest or unknowingly wrong — they simply make tradeoffs that sacrifice security or decentralization. Let’s take the easiest example: Binance Smart Chain. The argument that you could increase block size and decrease block time to achieve more scalability has been made since the Bitcoin Cash days. Here’s Elon Musk talking about it in 2021:
It seems intuitive — if Ethereum just tuned the parameters Elon alludes to, it’d be able to achieve orders of magnitude more scalability. Unfortunately, it’s not that simple. If you increased the block size, full nodes would need larger SSDs (solid state drives) to maintain greater state sizes, thus making it more out of reach for regular users to run nodes. If you couple this with faster block times, then these full nodes would not only have to maintain bloated state sizes, but also faster execution times, eating up the CPU and bandwidth of the underlying hardware! [4]. Binance Smart Chain did exactly this — they followed what Elon Musk advised, and consequently have 21 nodes on the network. Again, this is fine if a chain accepts the tradeoffs it makes, but it simply isn’t sustainable for the long-term decentralized vision of this industry.
Take a more complicated example: Solana. Solana extended the scope of Binance Smart Chain and said instead of changing the parameters of Ethereum to skew towards greater scalability, let’s create a blockchain from scratch that is solely optimized for better hardware and censorship resistance. In other words, nodes that run Solana would be required to have highly parallelized GPUs, large disks to store data, high bandwidth (1gps+ internet speed), and other parameters that make it infeasible for a normal user to run a Solana node using a regular computer [5]. You may think Solana is more centralized than Binance Smart Chain, but using objectively measurable statistics, this is not true. Here’s why:
Solana is a PoS network, meaning that “miners” on Solana don’t solve the inputs of complex cryptographic hash puzzles like Bitcoin, but instead put-up stake in the form of SOL tokens when validating so they are economically disincentivized to act dishonestly. Ideally, a blockchain has a fair distribution of stake across its validating nodes — the reason being is because in a PoS network like Solana, if a handful of nodes control 33% of the stake, they could censor the network and disrupt the flow of information. Anatoly, the founder of Solana, makes this following argument here:
What Solana does to maximize this property of censorship resistance (one component of decentralization) is it attempts to *maximize* the minimum set of validators needed to censor the network (33% of SOL staked for validating) — this is known as the Nakamoto coefficient. If it achieves this, then according to them, the network is sufficiently decentralized.
One problem that’s unaccounted for in Anatoly’s argument, however, is the difficulty in running a Solana node. Because of its hardware optimizations (parallelized transaction pipeline, proof of history, and other optimizations), users are incentivized to delegate their stake to validators, further centralizing the network. Further, even if the stake of validators were distributed across many, regular users wouldn’t be able to verify the state of the chain because they’re effectively blocked out from running a full node unless they install one in a local data center. This defeats the purpose of a blockchain — there should be a degree of plausible accountability (if users don’t agree with an update, or if they simply want to fork out because they deem the current chain too centralized, they should be able to without convincing validators that run-in data centers).
Avalanche and Luna, although occupying different design spaces, suffer similar problems. And we don’t even account for the economic sustainability of them! I won’t elaborate too much on this, but in essence, chains should ideally generate more transaction revenue than the inflation they pay out to validators — Solana’s validator payout to annualized transaction revenue ratio is extremely high, whereas Ethereum’s is sustainable. Obviously, the outcome of this is higher fees (which nobody wants), but it’s necessary to maintain economic sustainability, and not speculation-driven growth.
Modular Blockchains
So, you may be wondering how the industry goes about scaling blockchains in a credible way. Succinctly put, blockchains need to be modularized. We talked about how first-generation blockchains perform everything: data availability (users run a bloated full node), consensus (block production to achieve BFT, necessary for all blockchains), and execution (processing and validating transactions) simultaneously. For blockchains to scale to global ubiquity, they need to unbundle execution from consensus and data availability [6].
Through IBC (inter-blockchain communication protocol), Cosmos does this to a certain extent, but still trades off security [7]. Because a multi-chain world on Cosmos would mean that each blockchain has its own security model, if one blockchain were to be insecure, then it could affect all the other blockchains that route communication with each other — Polkadot suffers from a similar problem as well.
Thus, as of writing this article, there are two ways a blockchain can credibly scale while maintaining shared-security and decentralization: (1) centralized block production by virtue of sharding, or (2) centralized block production by virtue of other modular architectures. It is also possible for Solana to implement fraud proofs and data availability sampling so users can quickly verify the validity of the blockchain, but because of the inherent bottlenecks associated with a giant, globally synchronized state machine that doesn’t unbundle execution, off-chain execution (i.e. rollups) is orders of magnitude more efficient.
Centralized Block Production by Virtue of Sharding
This is the route Ethereum 2.0 is taking — for its end vision, there will be 64 (though you could have more) shards that optimize for data availability and consensus. In other words, because execution will be unbundled from consensus and data availability, the shards on Ethereum will provide a large surface area for rollups to dump their data on [8] [9]. Since the computation will be done off-chain, Eth 2.0 full nodes effectively need to only verify and run consensus on the data. Add in data availability sampling (allowing light clients to have the same assurances as full nodes through fancy math and cryptography, like erasure coding), and you could achieve 100,000–1,000,000 TPS, all while maintaining a decentralized and secure base layer (decentralized because light clients can verify the state of the chain, and secure because the shards have shared-security through inter-shard communication with the beacon chain). This article won’t cover the execution side of things, but you could image rollups that optimize for parallelized transaction processing and compression (basically all the execution Solana does and way more), while inheriting the decentralized properties that Ethereum offers on a base-level. With this being said, Ethereum 2.0 is still far from production ready, and there remains execution risk before it fulfills its vision.
Centralized Block Production by Virtue of Other Modular Architectures
Celestia is another layer 1 blockchain that adopted the modular blockchain paradigm, albeit taking a different approach than Ethereum 2.0. Similar to Eth 2.0, Celestia optimizes solely for execution off-chain and consensus and data availability on-chain. However, instead of doing sharding to provide a robust data availability layer (think the > surface area of data availability while maintaining decentralization, the better), Celestia would run Tendermint (same consensus as Cosmos) over its data availability layer, allowing it to interoperate with Cosmos. We discussed how Cosmos does not have shared security because of bottlenecks associated with its inter-blockchain communication protocol, but what if you can use Celestia as a consensus layer that Cosmos dumps its data on? This is what Celestia will do, allowing it to achieve shared security from its own validator set instead of individual zones on Cosmos, while leveraging the composability that Cosmos offers via IBC. Though there are associated nuances, you effectively fulfill the promise of sharding with this design [10].
Conclusion
All blockchains face a constrained maximization problem: there exists a finite capacity based on resource requirements of an end-user full node (i.e. BTC limited to a theoretical maximum 4MB per block). Optimizing block times/sizes and hardware requirements to lower latency and increase throughput on L1 is not feasible given the security and decentralization tradeoffs. Creating a mechanism for which the base layer offers the ability for multiple blockchains to communicate sacrifices security because of the inability to share security. However, modularity by virtue of centralized block production but decentralized block validation (data availability sampling and fraud proofs), coupled with off-chain execution, will allow blockchains to scale to global ubiquity.
References
[1] https://medium.com/international-blockchain-olympiad/blockchain-from-first-principles3d615701dce7
[2] https://medium.com/certik/the-blockchain-trilemma-decentralized-scalable-and-securee9d8c41a87b3
[3] https://coinmarketcap.com/alexandria/article/what-is-data-availability
[4] https://vitalik.ca/general/2021/05/23/scaling.html
[5] https://medium.com/solana-labs/7-innovations-that-make-solana-the-first-web-scaleblockchain-ddc50b1defda
[6] https://polynya.medium.com/rollups-data-availability-layers-modular-blockchainsintroductory-meta-post-5a1e7a60119
[7] https://docs.cosmos.network/master/ibc/overview.html
[8] https://vitalik.ca/general/2021/04/07/sharding.html
