When Everyone Goes Parallel, Somnia Goes Sequential

Key Takeaways

• The history of the internet is closely tied to speed. As internet speed increased, it enabled new forms of Web 2 applications, shaping today's Internet ecosystem. Similarly, blockchain must achieve significantly greater scalability than it currently has if it aims to support applications like the metaverse and AR/VR.

• Recent projects focused on scalability, such as Solana, Sui, MegaETH, and Fuel, have focused on parallel processing. While parallel processing does enhance scalability, it cannot address situations where multiple transactions simultaneously access the same state, such as popular NFT minting events, limiting its effectiveness.

  
  

• Unlike other projects, Somnia proposes maximizing the sequential processing of EVM transactions within a single core. To achieve this, Somnia converts EVM bytecode directly into native code executable by the CPU, leveraging parallel processing capabilities inherent in CPU hardware.

1. Background - Everyone Goes Parallel

1.1 The History of Internet Has Evolved Alongside Speed

Source: ByteByteGo

The history of the internet and its applications has evolved alongside improvements in speed. The TCP/IP standard and Tim Berners-Lee's World Wide Web established the foundation of the internet, initially enabling only limited data transfers, such as text-based emails and simple web browsing at speeds below 56 kbps.

From the mid-1990s, internet connections transitioned from traditional dial-up to DSL (Digital Subscriber Line) and cable modems, significantly improving speeds (1–10 Mbps). This advancement allowed various new services to emerge, including audio and video streaming, online gaming, and e-commerce.

In the 2000s, fiber optic technology was introduced, dramatically increasing internet speeds. Additionally, the development of mobile broadband infrastructure, such as 3G and 4G LTE, allowed high-definition streaming and OTT services on mobile devices (50 Mbps–1 Gbps).

Currently, gigabit internet is becoming increasingly common as fiber optic infrastructure expands worldwide. In some areas, connections exceeding 10 Gbps are already available, and mobile network speeds have accelerated dramatically with 5G, rivaling fiber optics. Consequently, applications such as ultra-high-definition streaming, cloud gaming, and IoT are now possible on a broader scale.

Despite these advancements, internet speeds still fall short in certain applications. For instance, cloud gaming has yet to consistently provide gamers with a fully satisfying user experience. As the AR/VR industry develops further, it will require extremely low latency and stable internet connectivity to support real-time interaction.

1.2 Blockchain Still Hungers for Scalability

Source: imagflip (@100y_eth)

Unlike the internet, which primarily evolved by emphasizing speed, the blockchain industry has diversified its growth by embracing various factors such as decentralization, censorship resistance, culture, and community. For example, though now collapsed, the Terra network succeeded in building an unprecedented community around the ideology of decentralized stablecoins. Meanwhile, projects like Abstract Chain are pursuing mass adoption through consumer-facing services (see “Abstract: A Blueprint For Disneyland In Crypto”).

Nevertheless, blockchain technology ultimately serves as infrastructure supporting diverse applications, much like the internet. Thus, numerous blockchain projects have continuously strived to improve scalability, leading to the emergence of new types of decentralized applications (dApps). Initially, networks like Polygon and BNB Chain appeared to address Ethereum’s limited scalability, providing users with faster environments suitable for DeFi. Later, high-performance L1 blockchains such as Solana and Sui, as well as Ethereum L2 solutions like Base and Arbitrum, enabled previously challenging services such as order-book DEX to flourish.

Source: Paradigm

However, compared to the centralized internet that people currently use, blockchain networks remain vastly slower. This is due to the fundamental nature of blockchain, which relies on decentralized servers, resulting in additional performance degradation from interactions between nodes, as well as constraints in hardware specifications to maintain decentralization. If blockchain aspires to become the next iteration of the internet, it must ultimately support various Web2 services (such as gaming, social networking, streaming, etc.) seamlessly. To achieve this, blockchain scalability must significantly improve beyond current capabilities.

1.3 Everyone Goes Parallel

Recently, numerous blockchain projects have proposed the concept of transaction parallelization as a solution for improving scalability. Many of these projects have identified the inherently sequential nature of Ethereum’s EVM transaction processing as the main reason for its low scalability. To address this, some have developed specialized VMs designed explicitly for parallel transaction execution—such as Solana, Sui, and Aptos—while others like Monad, MegaETH, and Polygon have introduced transaction parallelization into existing EVM frameworks.

A critical aspect of transaction parallelization is determining whether transactions access the same blockchain state. For instance, two transactions minting NFTs from the same contract affect the same state and therefore cannot be processed in parallel—they must be executed sequentially. Blockchain networks typically follow one of two approaches to check for potential transaction conflicts:

The first approach, called State Access, involves determining the states accessed by transactions before execution. Transactions that do not interact with the same state are processed in parallel, while conflicting transactions are executed sequentially. The second method, Optimistic Execution, initially executes all transactions simultaneously and subsequently reprocesses any conflicting transactions sequentially. Most blockchain projects emphasizing parallelization use one of these two methods, as illustrated by the following examples:

• Solana: Solana enables multi-threaded parallel processing through Sealevel. Transactions include structures called "Instructions," containing information about the states involved. Sealevel parallelizes transactions only when there is no overlapping state access. (See “Solana Mega Report V2 - Like Apple, but Unlike Apple”).

• Sui: Like Solana, Sui utilizes the State Access method for parallel transaction processing. Sui’s data model is object-centric, meaning that two transactions can execute in parallel if they don't simultaneously interact with the same object. (See “Complete Guide to Sui”).

• Aptos: Aptos leverages a parallel execution engine called Block-STM, a representative example of Optimistic Execution. It initially processes all transactions concurrently and then sequentially re-executes any conflicting transactions. Polygon and Sei have also adopted Aptos’ Block-STM for their parallel transaction systems.

• Monad: Monad similarly uses the Optimistic Execution method for transaction parallelization. (See “A Monster Combined With Narrative and Technology, Monad.”).

• MegaETH: MegaETH maximizes parallelization by applying parallel processing separately to block creation and block verification. During block creation, sequencers can utilize various Optimistic Execution techniques, including Block-STM. For block verification, it employs a stateless validation method, enabling quick parallel block validation.

• Fuel L2: Despite being a rollup within the Ethereum ecosystem, Fuel relies on its unique UTXO-based FuelVM. Consequently, transactions that do not reference the same UTXO can inherently execute in parallel, providing significant scalability advantages.

2. Takeaway - Somnia Goes Sequential

2.1 Does Parallelism Work?

Source: MegaETH

Is transaction parallelization truly innovative, or is it merely a buzzword leveraged by projects for marketing purposes? According to MegaETH's research, simulations combining Ethereum network blocks into batches of 1, 2, 5, and 10 for parallel processing demonstrated clear performance improvements as batch sizes increased.

Of course, since scalability doesn’t increase linearly with batch size, it’s hard to claim that parallelization offers dramatic benefits. This is largely due to the fact that transaction conflicts occur quite frequently in real blockchain environments.

Source: Somnia

The graph above, based on Ethereum data from 2023 analyzed by the Somnia team, shows contract addresses on the x-axis and the number of calls on the y-axis. As shown, a small number of contracts dominate the network in terms of call volume. This concentration implies a higher likelihood of transaction conflicts among those calling the same contracts.

Source: Somnia

This second graph, based on Ethereum data analyzed in June 2024 by the Somnia team, shows the frequency of receiver addresses for ERC-20 token transfers. Just like with smart contracts, token transfers are also highly concentrated, with certain addresses receiving a disproportionate number of transfers.

Source: Etherscan

Moreover, parallel processing often proves ineffective precisely when it's most needed—when network congestion skyrockets due to numerous related transactions interacting simultaneously. NFT minting events provide a prime example, notably Yuga Labs' Otherside NFT minting event.

At the peak of NFT popularity, Yuga Labs raised an impressive $450 million in seed funding, and everyone wanted to participate in their NFT mint. I myself once participated in minting, paying a gas fee of 1.93 ETH, equivalent to 5,500 Gwei. Given recent Ethereum network gas fees are typically below 1 Gwei, this was extraordinarily high (Ethereum was priced around $2,800 at the time).

Source: Somnia

In highly popular or financially rewarding NFT minting events, a large number of users interact simultaneously with the same NFT minting contract, generating minting transactions that affect the same blockchain state. Even blockchains capable of parallel processing become ineffective in such cases, as these transactions cannot be parallelized. In other words, parallel processing sometimes fails precisely when scalability is most critical.

To tackle this issue, Somnia proposes maximizing sequential transaction processing performance within a single core. Let’s briefly explore Somnia’s core features and understand how transactions can be processed rapidly on a single core.

2.2 Somnia’s Core Features

Somnia aims to achieve unprecedented blockchain scalability, enabling Web2 applications previously considered impossible in blockchain environments (e.g., on-chain metaverse and social applications). Somnia maximizes EVM performance through technologies including:

• Multistream Consensus: Unlike other blockchains, Somnia has each validator run its own unique blockchain called a Datachain. These datachains operate independently, and only the validator who owns a specific datachain can append blocks to it. Overall consensus is achieved via the Consensus Chain, which references the latest blocks from multiple datachains and determines their ordering. This decouples data generation and consensus, resulting in high efficiency.

• Sequential Execution: Somnia introduces its proprietary EVM compiler, leveraging hardware parallelism to rapidly execute transactions sequentially within a single CPU core.

• IceDB: Somnia utilizes a new database system called IceDB, improving gas fee efficiency and optimizing caching to significantly enhance read and write speeds.

• Advanced Compression Techniques: Somnia uses streaming compression instead of block-based compression to minimize redundant data transferred between nodes, thus facilitating rapid transaction processing.

These core features of Somnia will be discussed in greater depth in future articles. For this article, we focus on how Somnia optimizes sequential transaction processing.

2.3 How Somnia Improves Scalability

2.3.1 EVM Compilation

EVM fundamentally uses a stack-based architecture. In a stack-based structure, data is pushed onto a data structure called a stack, and commands are processed accordingly. Operations in stack-based systems have predefined operands (always using the data at the top of the stack), resulting in simplified bytecode. However, such an approach suffers from redundant and unnecessary operations due to repetitive pushing and popping of data.

Moreover, the EVM executes operations in an interpreted manner, simulating a virtual machine in software by decoding instructions one by one. This method leads to slow performance due to repeated instruction lookups and handling processes for each individual command.

To address this, Somnia introduces its own EVM compiler. This compiler directly translates EVM bytecode into native machine code executable by CPUs. The resulting compiled native code achieves performance similar to code directly written in C++.

However, converting EVM bytecode into native code requires significant computational resources. Therefore, Somnia efficiently manages resources by compiling only frequently called smart contracts into native code. In benchmark tests, Somnia achieved ~1.05M of TPS for ERC-20 token transfers.

2.3.2 Hardware-Level Parallelism

Instead of relying on software-level parallelization, Somnia leverages hardware-level parallelism to accelerate individual transaction execution speeds. Hardware-level parallelism involves executing multiple instructions simultaneously within a single CPU core to enhance performance. By translating EVM bytecode into native code, CPUs can internally perform instruction-level parallel execution. As a result, transaction executions are effectively parallelized, significantly improving performance.

For example, consider processing an ERC-20 token transfer. Typically, the EVM would sequentially perform:

1. Hashing the sender's account

2. Checking the sender's balance

3. Hashing the receiver's account

4. Checking the receiver's balance

However, when compiled to native code and executed with hardware-level parallelism, the sender operations (steps 1 and 2) and receiver operations (steps 3 and 4) can be processed simultaneously, effectively cutting total execution time in half.

2.4 Toward Truly On-Chain Applications

Many blockchain projects claim to improve scalability, but the scalability envisioned by Somnia differs significantly in scale. While other blockchain projects aim to support financial, gaming, or social applications, Somnia dreams of achieving true Web2-level scalability capable of supporting fully on-chain metaverse applications. To realize this ambitious goal, Somnia aims not only to innovate technically but also to provide a variety of features specifically designed for metaverse applications. Stay tuned for future articles that will dive deeper into Somnia’s groundbreaking innovations.

3. Resource

Related Articles, News, Tweets etc. :

• Somnia Docs

• All About Parallelization by Sui Foundation

• Solana Mega Report V2 - Like Apple, but Unlike Apple by Four Pillars

• Complete Guide to Sui by Four Pillars

• A Monster Combined With Narrative and Technology, Monad by Four Pillars