Breakpoint 2025: Anza Block

Nearly two years of uninterrupted network uptime, 200 billion transactions processed, and a roadmap that promises to transform Solana's market structure fundamentally—Anza's Brennan Watt delivered a comprehensive look at what's coming to the Solana protocol at Breakpoint 2025.

Summary

The presentation began with a powerful statistic: Solana has maintained solid green liveness for almost two years, processing around 200 billion transactions. This reliability hasn't come by accident—Anza's validator team has been running sophisticated attack simulations on the public testnet, cycling through different attack vectors every hour to stress-test the network's resilience.

What makes 2025 particularly exciting is the sheer volume of improvements shipping to mainnet. The network has already seen a 25% throughput improvement this year, with block limits increasing from 48 million to 60 million compute units. But the headline features are still coming: Alpenglow, a new consensus engine that slashes finality times to 150 milliseconds, and Multiple Concurrent Proposers (MCP), which will fundamentally change how blocks are built on Solana.

Anza isn't just focused on speed—developer and operator experience improvements are also in the pipeline. State costs are being reduced dramatically (currently around a million dollars per gigabyte), transaction sizes are expanding to 4KB, and validators will soon have protocol-native revenue sharing capabilities with stakers.

Key Points

Network Resiliency and Stress Testing

Anza's approach to network security has reached a level of sophistication that the team is only now comfortable discussing publicly. The validator team runs continuous attack simulations on the public testnet, with attacks triggering every hour. These fall into two main categories: network denial-of-service attacks that test load shedding and back pressure handling, and highly curated malicious blocks designed to stress the far corners of the Solana Virtual Machine.

The results speak for themselves. In one example shared during the presentation, a malicious node flooded other nodes with gigabits per second of network traffic—yet the network continued operating normally, producing blocks on time with tower height steady at 32. This evolution from "nodes would crash" to "network handles attacks without a hitch" represents years of paranoid, obsessive security work that enables Anza to ship aggressive releases with confidence.

Performance Breakthroughs with XDP

One of the most technically impressive achievements involves implementing kernel bypass for the network layer using XDP (eXpress Data Path). While blockchain discussions often cite network bandwidth as the bottleneck, Watt revealed that Solana's actual limitation was "slamming the sendmessage syscall, allocating memory in the kernel, and context switching."

XDP solves this by allowing network interface interaction directly from user space, resulting in a 200x throughput improvement for this performance bottleneck. The difference is so dramatic that the presentation's comparison chart required a logarithmic scale on the y-axis just to show the new latency alongside the legacy code. This feature is already available in Agave 3.0—Watt urged validators to enable it immediately to unlock the planned increase to 100 million compute units per block.

Doubling Hot Account Throughput

A critical but often overlooked improvement has been doubling the throughput to "hot accounts"—the most frequently accessed accounts on the network. These often represent the most popular assets or high-volume protocols. When Anza raised this limit from 12 to 24 million, users immediately consumed the additional capacity, demonstrating strong demand for this expanded headroom.

Transaction Ingestion Scaling

The network now routinely sustains 100,000 transactions per second ingestion on mainnet. Watt described this as measuring "how efficiently we can move bidders into the auction house"—a critical metric for market efficiency. High ingestion rates ensure that the highest bids actually win, which requires examining all submitted bids.

Upcoming changes will further improve ingestion: packet processing limits are increasing to one million per second, unstaked connections are jumping from 500 to 4,000, and per-connection packet limits will be unthrottled when global congestion is low.

Alpenglow Consensus Engine

Perhaps the most anticipated upgrade is Alpenglow, the new consensus engine replacing Tower. The benefits are substantial: transaction finality drops to just 150 milliseconds, safety guarantees increase, and the architecture enables more intelligent approaches to handling slot laggers.

The development team has been running a 50-node globally distributed test cluster under load for over four months with remarkably few issues. Feature completion and testnet deployment are targeted for early 2025.

Multiple Concurrent Proposers (MCP)

MCP represents a fundamental shift in how Solana blocks are built. The current single-builder model creates monopoly conditions—MCP ends this by enabling multiple validators to simultaneously propose transactions for inclusion in blocks.

The benefits flow to multiple stakeholders: users and applications gain optionality and better regional latency, traders get censorship resistance, and there's no middleman involved—validators simply play different roles in the process. Watt noted that skeptics predicted MCP would take two years, but features moving toward this vision will ship incrementally with each upcoming release.

Developer and Operator Experience

Anza is addressing long-standing community complaints about the cost of state on Solana. Currently, state storage costs approximately one million dollars per gigabyte on mainnet—dramatically out of proportion to actual storage costs (NVMe drives cost roughly 10 cents per gigabyte). Reducing this barrier will make it more affordable for projects to bootstrap with many users and accounts.

Transaction sizes are increasing from the current limit to 4KB, allowing more accounts to be touched without lookup tables and enabling more complex protocol operations. On the operations side, Anza is reducing disk wear, accelerating restarts by slowing incremental snapshot frequency, and implementing windcode instead of bincode for faster serialization.

Validator Revenue Sharing (SIMD-123)

The upcoming SIMD-123 implementation will provide protocol-native mechanisms for validators to share revenue with stakers and information providers. This matters because delegated stake directly increases the number of blocks a validator produces, and those blocks can be highly valuable. As Solana attracts increasing trading activity, fair profit-sharing with stakers becomes essential.

Facts + Figures

• Solana has maintained nearly two years of uninterrupted network liveness
• Approximately 200 billion transactions have been processed during this uptime period
• 25% throughput improvement achieved on Solana mainnet in 2024
• Block compute unit limits increased from 48 million to 60 million CUs
• Hot account throughput doubled from 12 to 24 million
• XDP implementation provides 200x throughput improvement for network operations
• Ingestion now regularly sustains 100,000 TPS on mainnet
• Unstaked connections increasing from 500 to 4,000
• Packet processing limits increasing to 1 million per second
• Alpenglow reduces transaction finality to 150 milliseconds
• A 50-node Alpenglow test cluster has run for over four months globally distributed under load
• Transaction size increasing to 4KB (up from current limits)
• Current state costs approximately $1 million per gigabyte on mainnet
• Anza's attack simulation team runs unique attack vectors every hour on public testnet
• 4x headroom exists on compute capacity
• Agave 3.1 release coming soon, followed by 4.0 with new features

Top Quotes

"Solana stays up. And it keeps servicing transactions."

"We used to be able to make nodes crash. Then we got a little better. Then we could disrupt nodes. Then we got a little more better. And now we can handle these things without a hitch."

"Solana showed everyone that blockchains can be fast and affordable. And now we're going to show them that they can be fair and highly efficient while remaining trustless."

"We're ending the single builder monopoly. Users and DAPs want optionality and regional network latencies. Traders want to have censorship resistance. And MCP is going to deliver all these things."

"Despite driven development is the most amazing thing ever. Thank you for fueling us."

"The rent is too damn high and the transactions are too damn small. And so we're going to fix this."

"At Anza, we don't do roadmaps. We do releases."

"The real alpha exists in plain sight on GitHub."

Questions Answered

What is causing Solana's improved reliability?

Anza has implemented an aggressive security testing regime where their validator team attacks the public testnet with unique attack vectors every hour. These attacks fall into two categories: network DOS attacks that stress load shedding and back pressure handling, and specially crafted malicious blocks that push the SVM to its limits. Through this continuous testing, they've progressed from a state where attacks could crash nodes to the current situation where the network handles sophisticated attacks without any disruption. This paranoid approach to security allows them to ship aggressive quarterly releases with confidence.

How is Solana achieving 25% throughput improvements?

The throughput gains come from addressing multiple resource bottlenecks systematically. Anza identified that they have approximately 4x headroom on compute and similar capacity on disk storage. Memory management has been significantly improved throughout the year. Most importantly, implementing XDP (eXpress Data Path) for the network layer provides a 200x improvement by bypassing kernel overhead. Block limits have increased from 48 million to 60 million compute units, and hot account throughput has doubled. These combined improvements translate into the 25% mainnet throughput increase.

What is Alpenglow and when is it coming?

Alpenglow is Solana's new consensus engine that will replace the current Tower consensus mechanism. It dramatically reduces transaction finality from the current timeframe down to just 150 milliseconds while also improving safety guarantees. The system has been running on a 50-node globally distributed test cluster under load for over four months with minimal issues. Anza plans to complete the feature and deploy it to testnet early in 2025. Beyond faster finality, Alpenglow provides the foundation for Multiple Concurrent Proposers and better approaches to handling validators who fall behind on slot production.

What is Multiple Concurrent Proposers (MCP) and why does it matter?

MCP fundamentally changes Solana's block building architecture by allowing multiple validators to simultaneously propose transactions for block inclusion, rather than having a single builder monopoly. This provides several benefits: users and applications get more options and better regional latencies, traders gain censorship resistance, and the system remains fully trustless with validators simply playing different roles. Anza is already building incrementally toward this vision, with features shipping in upcoming releases. This represents Solana's next evolution—proving blockchains can be fair and efficient while remaining decentralized.

Why is state storage so expensive on Solana and what's being done about it?

Current state storage costs approximately one million dollars per gigabyte on mainnet—dramatically disproportionate to actual hardware costs (NVMe drives cost about 10 cents per gigabyte). While Solana provides significant replication benefits, Anza acknowledges this pricing makes it difficult for projects to bootstrap with many users and accounts. They're working to lower these costs significantly, making the platform more accessible for new projects. The exact reduction targets weren't specified, but this is a priority focus area.

What improvements are coming for validators and operators?

Several operational improvements are in development. Disk wear is being reduced, and restarts are being made faster by slowing down incremental snapshot frequency and implementing windcode instead of bincode for faster serialization. SIMD-123 will provide protocol-native mechanisms for validators to share revenue from various sources with stakers and information providers. This matters because delegated stake directly increases block production, and as trading activity grows on Solana, fair profit-sharing becomes essential for the staking ecosystem.