Anza’s 2025 Mission: On the Road to 1M TPS

This article explores some of the key advancements that Agave has made over the last year and our roadmap for 2025. TL;DR – We are going to double block space this year and Agave is going to hit 1M TPS in the future. Increasing Bandwidth and Reducing Latency (IBRL) is essential for Solana to stay competitive. Complex systems are not optimized by one magic fix but rather the accumulation of thousands of micro-advancements. That’s why we are laser focused on accelerating improvements to the Agave client, which powers the Solana network.

Agave Today

As mainnet has adopted the v2.1 version of Agave, operators and users are noticing huge improvements. The skip rate - the percentage of time a slot is skipped due to a delinquent or slow leader node - has dropped below half a percent.

The slot times have become blazingly fast with the last few epochs averaging under 400ms slot times. Epoch 746 - the period ending Feb 24th, 2025 at 5:16 UTC - ripped through 432k slots in 47.5 hours for an average slot time under 396ms!

Transaction latency has similarly improved, with typical confirmation latencies under 1 second! TPS is consistently above the 4k mark.

Image from Validators.app Ping Thing

But we’re not done yet… the other line on the above chart shows the artificial constraint of 48 million CUs is consistently the limiting factor for bandwidth. The plan is to aggressively increase this cap, starting with a modest increase to 50M as part of SIMD-0207.

Why not just double the CU limit tomorrow? There are two main concerns that are actively being addressed:

1. Making sure the infra surrounding the chain is able to keep up. We’ve already observed CEXes and minimum spec’d RPC nodes struggling to keep up with increased TPS. If the write layer is drastically outpacing the read layer, the user experience degrades with it. To address this, we are actively IBRLing the surrounding infra in addition to the chain performance itself.

2. Delivering blocks to the cluster on time. The next largest bottleneck facing the Agave client is getting the block out from the leader node to the rest of the cluster. With current CU limits, this is not a problem, but as blocks grow significantly in terms of bytes, the median block distribution time can start exceeding 400ms for certain workloads. The cracked Anza devs are alleviating this bottleneck while also ensuring block packing is prioritized and constrained by the most critical resources.

Let’s dive into some of the technical details as to why Agave will be able to scale so aggressively this year.

Scaling Agave TPS: 2025 Focus Areas

In 2025, we are pushing Agave to handle higher transaction throughput and enable greater block capacity. Let’s dive into a few specific areas of optimization:

1. TPU Scheduler Optimization: The scheduler is a critical component of any validator client, as it determines the order and priority of transaction processing for block producers. Agave will implement several improvements in this area.

   1. More intelligent scheduling system to balance competing concerns such as: prioritizing highest fee payers, avoiding account collisions, and keeping worker threads loaded and busy executing. The new, greedy scheduler algorithm will reduce batching to ensure execution threads receive a steady flow of work. This better matches the real mainnet workload, which consists of heavier, swap type transactions as opposed to the small, simple transactions used in synthetic testing.

   2. Reduce memory allocations to avoid pipeline stalls when pagefaulting. Dynamic memory allocation can be very expensive, especially when the memory allocator tries to return memory back to the kernel using the madvise syscall. Reducing overall allocations helps minimize any pipeline stalls while returning this memory.

   3. Push certain filtering responsibilities up the pipeline so that a smaller, more highly curated flow of transactions enters the scheduler. Currently, filtering by priority is done in the scheduler, which is after other expensive operations such as signature verification. This means there is the potential to waste a lot of effort on low value packets that will never be scheduled. By filtering earlier, a ton of validator resources can be saved.

These improvements will not only increase throughput but also ensure that transactions are processed in a more efficient, timely, and fair manner.

2. Turbine Bandwidth Scaling: Turbine is Solana’s block propagation protocol, and its ability to deliver data quickly to validators is essential for performance. Agave will focus on improving its implementation of the turbine design by enhancing bandwidth, enabling faster block propagation across the network. By optimizing turbine performance, we can significantly reduce block propagation times and enhance overall network efficiency, especially during periods of high transaction volume.

One of the current issues is the time involved in computing the turbine tree, which uses a stake-weighted shuffle to determine which peers a given node needs to send to. A unique tree is computed for every single shred and must happen before a shred is sent to downstream nodes. By increasing the efficiency of this computation and speculatively computing the tree for future shreds during idle periods, the shred pipeline can flow more seamlessly and improve block times.

Another has to do with how efficiently data is provided to the Network Interface Card (NIC) for sending across the network. By reducing intermediate memory copies, a higher number of shreds sent per second can be achieved.

3. Status Cache Improvements: The status cache holds critical information regarding processed transactions and ensures that transactions are processed only once on a given fork. To improve scalability, Agave will introduce more efficient status cache management and locking strategies. This will allow for faster lookups and inserts and yield better overall validator performance.
   
   The current status cache uses global locks. The next version will make locking more granular by using a lock per blockhash/slot, which will enable more parallel access to the status cache.
   
   

4. Blockstore Insertion Enhancements: Blockstore shred insertion is a key aspect of blockchain data storage. A large part of this has to do with how block metadata is serialized, stored, deserialized, and fetched. Agave will implement optimizations that enable faster and more efficient insertion of blocks and metadata into the database, allowing the system to scale more effectively as transaction volumes increase.
   
   One example of this is how the blockstore tracks block completion. There is a shred index that is maintained in a B-tree to understand which shreds have been received for a given block. Serializing and Deserializing a B-tree containing thousands of indices gets expensive. By switching to a simple bitmap, SerDes time can be meaningfully reduced for this operation that is performed thousands of times per second.
   
   

5. Faster Hashing Algorithms: One of the simpler but more impactful optimizations is switching to a faster hashing algorithm. Agave is transitioning structures that don’t require more advanced protection to ahash instead of the default siphash. Ahash is designed for higher throughput and better performance, which will result in faster transaction processing and a more responsive blockchain. This change is expected to significantly reduce the time spent hashing and help achieve greater scalability.

Check out this Github project to see the full list of planned performance improvements and follow progress in real time!

Conclusion

Anza’s 2025 roadmap tracks our vision for Agave: make it the best client on Solana Mainnet. Working alongside Firedancer has also helped both teams improve their respective clients. The collaboration between the teams, which now involve many in person meetups per year, and deep coordination across all areas, have helped to unfurl and scrutinize every part of the Agave codebase. Healthy competition drives both teams to innovate and continuously push the envelope in optimizing performance.  By relentlessly tackling bottlenecks along the transaction and shred processing pipelines, Anza is set to tackle the challenges of 2025 and beyond, ensuring Agave and Solana remain fast, reliable, and ultra competitive.

Stay tuned for more updates as we continue to iterate on Agave to squeeze all the performance out of the hardware, unlock more block capacity, and enable an increasing number of use cases to come onchain!