TPS and What Comes Next

For years, the blockchain industry has chased TPS. To bring Web2-scale experiences on-chain, you still need high throughput and fast finality, and the space has tried a range of approaches: new consensus designs, parallel execution, sharding, and rollups. Yet while centralized Web2 platforms deliver real-time services to massive audiences, public chains have struggled to go fully on-chain under network surges, gas bidding wars, and lagging or dropped transactions.

The gap is most obvious in games, social, and AI, where interaction is high-frequency and real-time. Raising TPS alone doesn’t fix it: with correlated state, conflicts and retries creep in; during traffic spikes, finality and propagation can become inconsistent. Pushing only for bigger hardware skews the validator set and invites centralization risk. As interaction rates and user counts grow, the bar moves from being merely fast to staying predictably fast under load, backed by reliable operations. That reality has shaped new architectural approaches built for production, not just benchmarks.

Somnia: A High-Performance Blockchain for Real-Time Applications

Against this backdrop, Somnia arrives as a next-generation, EVM-compatible L1 aimed at fully real-time on-chain experiences — from large-scale games and AI-driven markets to social platforms and creator economies. Somnia advertises 1M+ TPS, sub-second finality, and sub-cent fees, and launched mainnet alongside record figures: over 2 billion transactions across six months of testing, and 80 million in a single day, an EVM record.

As a validator, DSRV ran in the testnet alongside 60+ peers with a focus on keeping the network steady during traffic spikes. On the strength of those results, Somnia moved to mainnet, bringing the same stability and TPS into production. Now, let’s dive into four core pieces that make it possible.

MultiStream Consensus

Drawing on ideas from Autobahn BFT, MultiStream assigns each validator its own data stream (chain), while a separate consensus layer finalizes the stream heads. Instead of funneling everything into a single queue, it’s multiple checkout lanes with a final reconciler; you avoid global bottlenecks and keep latency more predictable as volume rises.

• Consensus efficiency holds up as throughput grows

• Streamed data reduces communication overhead

• Throughput scales without concentrating block production

Accelerated Sequential Execution

Rather than forcing more parallelism where transactions collide, Somnia makes the single lane extremely fast. Somnia compiles EVM bytecode to optimized native code, squeezing hardware parallelism (CPU cache, pipelining) within a single thread. The result is short, steady latency even with interdependent transactions — and it stays simpler to implement and debug.

• Native compilation maximizes execution speed

• Avoids sync overhead of naive parallelism on correlated state

• Simpler to implement and debug in production

IceDB (Purpose-Built State Database)

IceDB is Somnia’s custom state database and a core part of its transaction pipeline. It measures the real cost of reads and writes; in-memory reads are on the order of 15–100 ns, and snapshot commits persist state without paying heavy Merkle-tree update costs. With these measurements, gas can be priced consistently and rationally, helping the network maintain stable throughput even under load.

• 15–100 ns in-memory reads

• Snapshot commits instead of heavy Merkle updates

• Precise metering of RAM/disk access for consistent gas pricing

Advanced Compression & BLS Aggregation

Treating validator data as continuous streams enables compression over recent history rather than block-by-block isolation. In addition, BLS signature aggregation collapses thousands of signatures into a single signature per batch. The net effect: with the same network and hardware, you process more transactions, reduce latency, and scale without shrinking the validator set.

• Stream-aware compression cuts redundant bytes

• BLS aggregation replaces thousands of signatures with one

• Pushes toward physical bandwidth limits without centralizing the network

  
  

From the Validator’s Console

The most tangible advantage of Somnia’s core tech is predictable latency under load. MultiStream prevents a single global queue from clogging; accelerated sequential execution keeps sync overhead low even with correlated transactions. Because IceDB meters read/write cost, we can quickly localize delays, which speeds incident response and simplifies capacity planning. With solid bandwidth and low link latency, stream-based compression and BLS aggregation materially reduce propagation burden.

As part of the testnet, DSRV provisioned its nodes above published specs, allocating extra RAM for IceDB’s memory profile; that headroom helped us ride traffic spikes smoothly. In practice, block times and transaction processing stayed stable regardless of traffic, which is an important trust signal from an operator’s perspective. Looking ahead to mainnet scale, clear guidance and best practices for disk-usage optimization would help operators plan efficiently, as growing throughput and state make operational efficiency and cost control more important.

During testnet, we handled multiple upgrades, migrations, and operational events typical of a testnet, all of which went smoothly thanks to the Somnia team’s responsive support and real-time docs/guides. Collaboration within the validator community was also notable — open exchanges during upgrades or incidents helped both operational stability and community trust.

Ecosystem Momentum

Diverse Real-World Use Cases

What stands out beyond TPS is how quickly actual use cases are landing on the stack. Projects across gaming (e.g., Sparkball, Variance, Maelstrom), creator platforms, and financial apps are entering the ecosystem.

• Gaming: Real-time matchmaking, combat, and item economies need on-chain rules with a Web2-smooth feel — predictable latency is the quality bar.

• AI agents/markets: Agents must record on-chain and react immediately — short round trips and steady processing are critical.

• Social/creator: Bot resistance and automated rights/revenue sharing require on-chain enforcement without degrading UX.

Builder Onboarding

Somnia reduces onboarding and ops friction with Somnia Builder and SDKs (Unity, Unreal, TypeScript, mobile), plus social login and gasless flows. Integrations with LayerZero, Sequence, Ankr, DIA, Thirdweb bring familiar tools and interoperability with the broader Web3 stack.

Enterprise-Grade Confidence

Google Cloud’s participation brings real-time data streaming (Pub/Sub), BigQuery indexing and analytics, and Mandiant threat intelligence meeting the visibility, security, and compliance needs of large-scale projects. If Google’s AI agent ecosystem is integrated, it could enable features like dynamic NPCs, AI-driven market participants, and reactive in-game mechanics that respond instantly to player actions.

Conclusion

As new games and applications come online, the number of transactions and state transitions to validate explodes, making stable consensus, low-latency propagation, and predictable gas non-negotiable. In that context, the validator’s role is straightforward. DSRV continuously monitors uptime, latency, failure rates, and propagation times, focusing on prevention and capacity planning to keep the network dependable.

Post-launch, Somnia aims to roll out performance monitoring/optimization tools and broaden the validator set. Following its successful mainnet debut, Somnia appears well placed to become a foundational infrastructure layer, delivering the scalability and reliability required for next-generation on-chain applications.