Towards the World's Supercomputer: A New Paradigm for Decentralized Execution at Hyperscale

To achieve decentralization, take advantage of the inherent trustlessness of cryptography, the natural economic incentives of MEV to drive mass adoption, the potential of ZK technology, and the need for decentralized general-purpose computing, including machine learning, the emergence of the world's supercomputers has become necessary.

Original title: "Towards World Supercomputer"

Written by: msfew, Kartin, Xiaohang Yu, Qi Zhou

Compilation: Deep Tide TechFlow

introduce

How close is Ethereum to eventually becoming that world's supercomputer?

From Bitcoin's peer-to-peer consensus algorithm to Ethereum's EVM to the concept of a network nation, one of the goals of the blockchain community has always been to build a world supercomputer, more specifically, a decentralized, unstoppable , trustless and scalable unified state machine.

While it has long been known that all of this is very theoretically possible, most ongoing efforts to date have been very fragmented and have serious trade-offs and limitations.

In this article, we explore some of the trade-offs and limitations faced by existing attempts to build a world computer, then analyze the components necessary for such a machine, and finally propose a novel world supercomputer architecture.

A new possibility, worthy of our understanding.

1. Limitations of the current method

a) Ethereum and L2 Rollups

Ethereum was the first real attempt at building the world's supercomputer, and arguably the most successful. However, during its development, Ethereum greatly prioritized decentralization and security over scalability and performance. So, while reliable, regular Ethereum is far from being the world's supercomputer - it simply doesn't scale.

The current solution is L2 Rollups, which have become the most widely adopted scaling solution for enhancing the performance of computers in the Ethereum world. As an additional layer built on top of Ethereum, L2 Rollups offer significant advantages and are supported by the community.

While multiple definitions of L2 rollups exist, it is generally accepted that L2 Rollups are networks with two key characteristics: on-chain data availability and off-chain transaction execution on Ethereum or other underlying networks. Basically, historical state or input transaction data is publicly accessible and committed to verification on Ethereum, but all individual transactions and state transitions are moved off the mainnet.

While L2 Rollups have indeed greatly improved the performance of these "global computers", many of them have a systemic risk of centralization, which fundamentally undermines the principles of the blockchain as a decentralized network. This is because off-chain execution involves not only individual state transitions, but also the sequencing or batching of those transactions. In most cases, the L2 orderer does the ordering, while the L2 validators compute the new state. However, providing this ordering capability to L2 orderers creates a centralization risk, where the centralized orderer can abuse its power to arbitrarily censor transactions, disrupt network liveness, and profit from MEV capture.

Although there have been many discussions on ways to reduce the risk of L2 centralization, such as through sharing, outsourcing or orderer-based solutions, decentralized orderer solutions (such as PoA, PoS leader selection, MEV auction, and PoE), among them Many attempts are still in the conceptual design stage and are far from being a panacea for this problem. Additionally, many L2 projects seem reluctant to implement a decentralized sorter solution. For example, Arbitrum proposes a decentralized sorter as an optional feature. In addition to the centralized orderer issue, L2 Rollup may have centralization issues from full node hardware requirements, governance risk, and application rollup trends.

b) L2 Rollups and the World Computer Trilemma

All these centralization issues that come with relying on L2 Rollups to scale Ethereum reveal a fundamental problem, the "world computer trilemma", which is derived from the classic blockchain "trilemma":

Different priorities for this trilemma will result in different trade-offs:

• Strong consensus ledger: essentially requires repeated storage and calculation, so it is not suitable for expanding storage and calculation.
• Strong computing power: It is necessary to reuse the consensus when performing a large number of computing and proof tasks, so it is not suitable for large-scale storage.
• Strong storage capacity: It is necessary to reuse the consensus when performing frequent random sampling space proofs, so it is not suitable for computing.

The traditional L2 scheme is actually to build the world computer in a modular way. However, since the different functions are not partitioned based on the aforementioned priorities, the World Computer maintains Ethereum's original mainframe architecture even with scaling. This architecture cannot satisfy other functions such as decentralization and performance, and cannot solve the trilemma of the world computer.

In other words, L2 Rollups actually implement the following functions:

• Modularization of the world computer (more experiments on the consensus layer and some external trust on the centralized orderer);
• World Computer throughput enhancements (though not strictly "expanded");
• Open innovation of the world computer.

However, L2 Rollups do not provide:

• Decentralization of the world computer;
• Performance enhancement of the world computer (Rollups' max TPS combined is actually not enough, and L2 cannot possibly have faster finality than L1);
• Computation by the World Computer (this involves computation beyond transaction processing, such as machine learning and oracles).

While the world computer architecture can have L2 and modular blockchains, it doesn't solve the fundamental problem. L2 can solve the blockchain trilemma, but not the trilemma of the world computer itself. So, as we have seen, current approaches are not enough to truly realize the decentralized world supercomputer that Ethereum originally envisioned. We need performance expansion and decentralization, not performance expansion and gradual decentralization.

2. Design goals of the world's supercomputers

For this, we need a network that can solve truly general-purpose intensive computations (especially machine learning and oracles), while retaining the full decentralization of the base-layer blockchain. Additionally, we must ensure that the network is capable of supporting intensive computations, such as machine learning (ML), that can be run directly on the network and ultimately verified on the blockchain. In addition, we need to provide sufficient storage and computing power on top of existing world computer implementations, the goals and design methods are as follows:

a) Calculation requirements

To meet the needs and purposes of a world computer, we extend the concept of a world computer described by Ethereum and aim to achieve a world supercomputer.

The world's supercomputer first needs to complete the tasks that computers can complete now and in the future in a decentralized manner. To prepare for mass adoption, developers need the world's supercomputers to accelerate the development and adoption of decentralized machine learning to run model inference and validation.

For computing resource-intensive tasks like machine learning, achieving such a goal requires not only trust-minimizing computing techniques such as zero-knowledge proofs, but also greater data capacity on the decentralized network. These cannot be achieved on a single P2P network (such as traditional blockchain).

b) Solutions to performance bottlenecks

In the early days of computing, our pioneers faced similar performance bottlenecks as they made trade-offs between computing power and storage capacity. Take the smallest component of a circuit as an example.

We can compare computation to a light bulb/transistor and storage to a capacitor. In a circuit, a light bulb requires an electric current to emit light, similar to a computational task requiring computation to perform. Capacitors, on the other hand, store charge, similar to how storage can store data.

For the same voltage and current, there may be a trade-off in energy distribution between the bulb and capacitor. Typically, higher computations require more current to perform the computation task and therefore require less energy to be stored by the capacitor. Larger capacitors can store more energy, but may result in lower computational performance at higher computational loads. This trade-off makes it impossible to combine compute and storage in some cases.

In the von Neumann computer architecture, it led to the concept of separating the storage device from the central processing unit. Similar to separating the light bulb from the capacitor, this could solve the performance bottleneck of our world's supercomputer systems.

In addition, traditional high-performance distributed databases adopt a design scheme that separates storage and computing. This scheme was adopted because it is fully compatible with the characteristics of the world's supercomputers.

c) Novel architecture topology

The main difference between modular blockchains (including L2 Rollups) and world computer architectures is their purpose:

• Modular blockchain: Aims to create new blockchains by selecting modules (consensus, data availability layer DA, settlement and execution) and combining them into modular blockchains.
• World Supercomputer: Aims to build a global decentralized computer/network by combining networks (base layer blockchain, storage network, computing network) into a world computer.

We propose an alternative that the eventual world supercomputer will consist of three topologically heterogeneous P2P networks connected by trustless buses (connectors) such as zero-knowledge proof technology: consensus ledger, computing network, and storage network. This basic setup enables the world's supercomputers to solve the world's computer trilemma, and other components can be added as needed for a particular application.

It is worth noting that topological heterogeneity involves not only architectural and structural differences, but also fundamental differences in topological forms. For example, while Ethereum and Cosmos are heterogeneous in terms of network layers and interconnections, they are still equivalent in terms of topological heterogeneity (blockchains).

In the world's supercomputers, the consensus ledger blockchain adopts the form of blockchain, and the nodes adopt the form of a complete graph, while the zkOracle network like Hyper Oracle is a network without ledgers, and the nodes form a cyclic graph, while the network structure for storing Rollup is Another variant, partitions form subnets.

By using zero-knowledge proofs as the data bus, we can achieve a fully decentralized, unstoppable, permissionless and scalable world supercomputer by connecting three topologically heterogeneous peer-to-peer networks.

3. World Supercomputer Architecture

Similar to building a physical computer, we must assemble the previously mentioned consensus network, computing network, and storage network into a world supercomputer.

Proper selection and connection of each component will help us achieve a balance between the consensus ledger, computing power and storage capacity trilemma, and ultimately ensure the decentralization, high performance and security of the world's supercomputers.

The architecture of the world's supercomputers is described as follows according to their functions:

The node structure of a world supercomputer network with consensus, computation and storage network resembles the following:

To launch the network, the nodes of the world's supercomputer will be based on Ethereum's decentralized infrastructure. Nodes with high computing performance can join zkOracle's computing network for generating proofs for general computing or machine learning, while nodes with high storage capacity can join EthStorage's storage network.

The above examples describe nodes running both Ethereum and compute/storage networks. For nodes that only run computing/storage networks, they can access Ethereum's latest blocks or prove the availability of stored data through a bus of zero-knowledge proof technologies such as zkPoS and zkNoSQL without trust.

a) Ethereum Consensus

Currently, the consensus network of the world's supercomputers exclusively uses Ethereum. Ethereum has strong social consensus and network-level security, ensuring decentralized consensus.

The world's supercomputers are built on a consensus ledger-centric architecture. The consensus ledger serves two main purposes:

• Provide consensus for the entire system;
• Define the CPU clock cycle with the block interval.

Compared with computing networks or storage networks, Ethereum cannot handle a large number of computing tasks at the same time, nor can it store large amounts of general-purpose data.

Among the world's supercomputers, Ethereum is a consensus network for storing data, such as L2 Rollup, to reach a consensus for the computing and storage network, and to load key data so that the computing network can perform further off-chain calculations.

b) Store Rollup

Ethereum's Proto-danksharding and Danksharding are essentially ways to scale the consensus network. In order to achieve the storage capacity required by the world's supercomputers, we need a solution that is both native to Ethereum and supports permanent storage of large amounts of data.

Storage Rollups, like EthStorage, essentially scale Ethereum for massive storage. Also, since computationally resource-intensive applications such as machine learning require large amounts of memory to run on physical computers, it is important to note that Ethereum's "memory" cannot be overscaled. Storage Rollups are necessary for the "swapping" that allows the world's supercomputers to run computationally intensive tasks.

In addition, EthStorage provides a web3:// access protocol (ERC-4804), which is similar to the native URI or storage resource addressing of the world's supercomputers.

c) zkOracle computing network

The computing network is the most important element of the world's supercomputers because it determines the overall performance. It must be able to handle complex computations such as oracles or machine learning, and should be faster than consensus and storage networks in terms of accessing and processing data.

The zkOracle network is a decentralized and trust-minimized computing network capable of processing arbitrary computations. Any running program generates a ZK proof, which when used can be easily verified by consensus (Ethereum) or other components.

Hyper Oracle is a network of zkOracles, powered by zkWASM and EZKL, that can run any computation using proof-of-execution traces.

The zkOracle network is a ledgerless blockchain (no global state) that follows the chain structure of the original blockchain (Ethereum), but operates as a computing network without ledgers. The zkOracle network does not guarantee computational validity through re-execution like traditional blockchains; instead, it provides computational verifiability through generated proofs. The ledger-less design and dedicated node setup for computing allows zkOracle networks (such as Hyper Oracle) to focus on high-performance and trust-minimized computing. The result of calculation is directly output to the consensus network instead of generating a new consensus.

In zkOracle's computing network, each computing unit or executable file is represented by a zkGraph. These zkGraphs define computation and proof generation behavior, just like smart contracts define the computation of a consensus network.

I. General Off-Chain Computing

The zkGraph program in zkOracle's computation can be used without an external stack for two main use cases:

• indexing (accessing blockchain data);
• Automation (automated smart contract calls);
• Any other off-chain computation.

These two cases can meet the middleware and infrastructure requirements of any smart contract developer. This means that as a developer of the world's supercomputer, you can go through the entire end-to-end decentralized development process when creating a complete decentralized application, including smart contracts on the chain on the consensus network and chains on the computing network. Next calculate.

II. ML / AI calculations

To achieve Internet-scale adoption and support any application scenario, the world's supercomputers need to support machine learning computing in a decentralized manner.

Through zero-knowledge proof technology, machine learning and artificial intelligence can be integrated into the world's supercomputers and verified on Ethereum's consensus network to achieve real on-chain computing.

In this case, zkGraph can be connected to external technology stacks, thus combining zkML itself with the computing network of the world's supercomputers. This enables all types of zkML applications to:

• ML/AI for user privacy protection;
• ML/AI for model privacy protection;
• ML/AI with computational efficiency.

In order to achieve the machine learning and artificial intelligence computing power of the world's supercomputers, zkGraph will be combined with the following advanced zkML technology stacks, providing them with direct integration with consensus networks and storage networks.

• EZKL: Perform inference in zk-snark for deep learning models and other computational graphs.
• Remainder: Fast machine learning operations in Halo2 Prover.
• circomlib-ml: circom circuit library for machine learning.

e) zk as data bus

Now that we have all the basic components of the world's supercomputer, we need a final component to connect them. We need a verifiable and trust-minimized bus to communicate and coordinate between components.

Hyper Oracle zkPoS is a suitable candidate for the zk Bus for the world's supercomputers using Ethereum as a consensus network. zkPoS is a key component of zkOracle, which verifies the consensus of Ethereum through ZK, so that the consensus of Ethereum can be propagated and verified in any environment.

As a decentralized and trust-minimized bus, zkPoS can connect all components of the world's supercomputers through ZK, with almost no verification calculation overhead. As long as there is a bus like zkPoS, data can flow freely within the world's supercomputers.

When the consensus of Ethereum can be passed from the consensus layer to the bus as the initial consensus data of the world's supercomputers, zkPoS can prove it through state/event/transaction proof. The generated data can then be passed to the computing network of the zkOracle network.

Additionally, for the storage network's bus, EthStorage is developing zkNoSQL to enable proofs of data availability, allowing other networks to quickly verify that a BLOB has sufficient replicas.

f) Another case: Bitcoin as a consensus network

Like many second-layer sovereign rollups, a decentralized network like Bitcoin can serve as the consensus network underpinning the world's supercomputers.

In order to support such a world supercomputer, we need to replace the zkPoS bus, because Bitcoin is a blockchain network based on the PoW mechanism.

We can use ZeroSync to implement zk as the bus of the Bitcoin world supercomputer. ZeroSync is similar to "zkPoW", which synchronizes Bitcoin's consensus through zero-knowledge proofs, allowing any computing environment to verify and obtain the latest Bitcoin status within milliseconds.

g) Workflow

The following is an overview of the transaction process of the world's supercomputer based on Ethereum, broken down into several steps:

• Consensus: Use Ethereum to process and reach transaction consensus.
• Computation: The zkOracle network performs relevant off-chain computations (defined by zkGraph loaded from EthStorage) by quickly verifying proofs and consensus data delivered by zkPoS as a bus.
• Consensus: In some cases, such as automation and machine learning, the computing network will pass data and transactions back to Ethereum or EthStorage through proofs.
• Storage: For storing large amounts of data from Ethereum (such as NFT metadata), zkPoS acts as a messenger between Ethereum smart contracts and EthStorage.

Throughout the process, the bus plays a vital role in connecting each step:

• When consensus data is passed from Ethereum to the computing of the zkOracle network or the storage of EthStorage, zkPoS and state/event/transaction proofs generate proofs that the receiver can quickly verify to obtain the exact data, such as the corresponding transaction.
• When the zkOracle network needs to load data from storage for calculation, it uses zkPoS to access the address of the data from the consensus network, and then uses zkNoSQL to get the actual data from storage.
• When data from the zkOracle network or Ethereum needs to be displayed in the final output form, zkPoS generates proofs for clients (such as browsers) for fast verification.

in conclusion

Bitcoin has laid a solid foundation for the creation of the world computer v0 and successfully built the "world ledger". Subsequently, Ethereum effectively demonstrated the "world computer" paradigm by introducing a more programmable smart contract mechanism. To achieve decentralization, taking advantage of the inherent trustlessness of cryptography, the natural economic incentives of MEV, driving mass adoption, the potential of ZK technology, and the need for decentralized general-purpose computing, including machine learning, the emergence of the world's supercomputers has become necessary.

Our proposed solution will build a world supercomputer by connecting topologically heterogeneous P2P networks using zero-knowledge proofs. As a consensus ledger, Ethereum will provide the basic consensus and use the block interval as the clock cycle of the entire system. As a storage network, a storage rollup will store large amounts of data and provide URI standards to access the data. As a computing network, the zkOracle network will run resource-intensive computations and generate verifiable proofs of computation. As a data bus, zero-knowledge proof technology will connect various components and allow data and consensus to be linked and verified.