Sub-Second Latency: Aptos Delivers Instant Transactions
By Brian Cho, Manu Dhundi, Sital Kedia, Zekun Li, Alexander Spiegelman, Satya Vusirikala
Editor’s note: This report was modified on May 22, 2024 and will continue to evolve as others in the industry contribute to our open-source benchmarks. We encourage everyone to submit PRs to continue providing the industry with fair and consistent latency benchmarks.
TLDR: Aptos is revolutionizing the blockchain experience by being the first to deliver consistent sub-second E2E latency for all transaction types.
In the fast-paced world of digital transactions, timing is everything. Whether you’re swiftly swapping assets on a decentralized exchange (DEX) or immersed in a thrilling RPG, waiting for blockchain confirmations can be a real buzzkill. Even mundane tasks like paying for a coffee shouldn’t come with a side order of waiting. In short, low end-to-end transaction latencies are paramount for a great Web3 user experience.
End-to-End (E2E) Latency
E2E Latency is the duration for a transaction to be submitted by a user and acknowledged as irrevocably settled (or finalized) on a blockchain. This includes the time it takes for the transaction to reach the validators, achieve consensus, validate and execute the transaction, ensure reliable storage, and provide the transaction confirmation back to the user.
E2E Latency represents true user experience while interacting with the blockchain, distinct from internal measurements like ‘blocktime’ that can be misleading. Transaction confirmation on Aptos guarantees irrevocable finality, which is stronger than other forms of finality like probabilistic and optimistic finality provided by some blockchains.
E2E Latency Comparison Benchmark
At Aptos, we recognize that verifiable and repeatable benchmarks fostering fair comparisons are essential for the entire blockchain industry to thrive. Following the success of our 2023 throughput benchmark, we’re excited to unveil a new benchmark to measure E2E Latency across blockchains. The numbers produced by the benchmark represent an end user’s observed latency from submitting a transaction to getting a transaction confirmation back.
We compared the latency of coin transfer transactions across multiple blockchain networks in their mainnet environments. Coin transfer is a straightforward and cost-effective transaction type that is widely supported by SDKs.
The benchmark is a job running from GCP-Tokyo, periodically transferring a small amount of coins between a predefined sender and receiver account. It was run from Asia — where most Web3 users are today — and leverages JavaScript SDKs for the respective blockchains with a frequency of 15 minutes to avoid overloading any blockchain and keep the cost of experimentation low. We use publicly available RPC endpoints and pay the default gas fees.
We make a comparison across several chains that are intended for high-performance usage: Aptos, Solana, Near, Avalanche (C chain), Sui, Arbitrum, Optimism, Base and Polygon. More details on the benchmark and how to run it can be found here. The real time results from the benchmark can be viewed here.
As the graph above shows, Aptos is the fastest blockchain to consistently achieve sub-second E2E latency for all transaction types. It’s worth noting that even though the above benchmark was run in Asia, Aptos achieves sub-second latency across all other tested regions as well.
Aptos: Engineered for Speed
A transaction’s journey takes it through the entire blockchain: the transaction is created at the client/SDK, submitted to a mempool on a fullnode, propagated to the validator set where it is ordered in consensus and executed, then returned back to the fullnode through state sync, and finally the result is read at the client/SDK.
To get sub-second latency in a decentralized blockchain environment is challenging due to networking and speed-of-light constraints. Client/SDKs and fullnodes are distributed throughout the world. A geo-distributed decentralized validator set requires average 100+ ms latency to send messages across the set, making fast finality a challenge.
To reduce latency, we rely on rock-solid architecture and optimizations at each part of the transaction’s path:
Mempool
The mempool is the staging area for incoming transactions; the mempool validates and forwards transactions upstream to the validator set where they are inserted into the Quorum Store. The mempool’s primary function is to queue and prioritize transactions under load and is implemented across fullnodes and validators to safeguard against Distributed Denial of Service (DDoS) attacks. The mempool is gas-aware, to prioritize the highest gas transactions first when propagating upstream.
To achieve sub-second latency the mempool is optimized to achieve fast propagation along fullnodes to the validator set. Recently fullnodes have been upgraded to do latency-aware peer selection. They ping upstream validator full nodes to measure latency, and select a set of upstream nodes with both the lowest latency and proximity to the validator set.
Lightning Fast Byzantine Fault Tolerance (BFT) Consensus
The validator set provides a replicated state machine where user transactions are executed, by running Quorum Store, BFT Consensus, and Execution and Commit of blocks.
The BFT consensus protocol, AptosBFT v4 is based on Jolteon and is provably resilient to a third of validator stake failing (including malicious behavior). On this foundation, we have implemented a new set of optimizations that reduces consensus latency from 5 to 3 message delays. This brings Aptos consensus latency in line with the theoretical lower bound, establishing Aptos as the fastest blockchain in the industry.
To further optimize for sub-second latency, we implemented adaptive block proposals. Quorum Store’s key purpose is to increase throughput by having proposals include signed quorum certificates of batches of transactions instead of raw transactions, which can be large for high numbers of transactions. With adaptive block proposals, proposers can include inline batches that are yet to receive quorum certificates. This speeds up the time transactions are added to proposals.
High Throughput Enables Low Latency Under Heavy Load
Our software stack is meticulously crafted to uphold unmatched throughput, ensuring our platform remains responsive even during peak demand. In a recent experiment conducted on our previewnet — a testbed akin to our mainnet — we achieved a staggering peak throughput of 30,000 transactions per second (TPS) and seamlessly processed an unprecedented 2 billion transactions in a single day.
Behind this impressive performance lie several key innovations unique to Aptos. Our horizontally scalable Quorum Store enables consensus to effortlessly scale to meet increasing demands. Dynamic parallelism, facilitated by BlockSTM, ensures transactions are processed concurrently, boosting execution throughput dramatically. Moreover, our highly optimized batched storage solutions minimize storage overhead, while pipelined processing of stages optimizes resource utilization across the board. Together, these innovations form a robust framework that empowers our blockchain to maintain low latency even under the most demanding conditions.
Streaming state synchronization
Transaction results are sent back downstream via streaming state sync — a feature designed to minimize user-perceived latency by streaming transaction confirmations to downstream nodes with minimal latency. Leveraging our latency-aware peer selection mechanism, downstream peers intelligently select the nearest and most reliable upstream peers. Once connected, downstream peers subscribe to transaction confirmation notifications. When an upstream peer receives a transaction confirmation, it immediately streams it downstream without any additional polling delay.
Observations Beyond Raw Numbers
Block Time
The E2E Latency should not be confused with ‘Block Time,’ which refers to the duration required for a blockchain to produce a new block. While Block Time is a metric to measure blockchain performance, it falls short of providing a full picture of end to end latency because it refers to the duration of a subset of stages of transaction processing — not what users actually experience.
Types of Finality
When Aptos transactions are confirmed to the user, they are already written to the blockchain with a guarantee of being irrevocably final. Some blockchains reduce end-to-end latency by providing weaker finality guarantees. Broadly speaking, there are 3 types of finality:
Probabilistic Finality — In certain blockchains (e.g., Bitcoin), as time passes the probability of a block being removed from the blockchain due to reorganization of blocks reduces, and only after a certain threshold it is close to zero, and then block is considered to be final (e.g., after tens of minutes).
Optimistic Finality — Certain blockchains (like Solana) expose an early confirmation of the block based on certain rules like gathering enough votes from the validators to provide certain finality guarantees — however, these finality guarantees are not irrevocable. For example, in case of Solana, dishonest validators (that require stake slashing) can cause an optimistically confirmed transaction to be rolled back thereby violating the assumed finality. Therefore a transaction is deemed irrevocable only after 31+ confirmed blocks are built on top of the optimistic confirmation. From our measurements we can see that while optimistic confirmation is achieved by Solana in around 5s, irrevocable finality (in the way guaranteed by Aptos) is achieved after around 25s.
Irrevocable Finality — This is the strongest form of finality, which guarantees that once a block is finalized, it is not possible to revert it as long as BFT assumptions hold (not more than 33% of the stake is malicious). Aptos provides irrevocable finality to guarantee the highest level of safety to the users. Once a transaction is confirmed, it will not be revoked.
Latency Based on The Transaction Type
In the majority of chains evaluated in the benchmark, the end-to-end latencies observed with coin transfer transactions provide an indication of the latencies associated with all transactions. In essence, a coin transfer transaction undergoes the same path as any other typical transaction, without receiving any preferential treatment. However, Sui blockchain stands out as an exception in this context. Sui employs a “fast path”, specifically for limited types of transactions (like coin transfer transactions) while the remaining follow a regular consensus path.
L2 Promise: But, Where’s The Finality?
Several L2 solutions are built to address scalability issues on L1 blockchains like Ethereum. Popular L2s like Arbitrum, Optimism, Base, and Polygon aim to make transactions on Ethereum faster and cheaper. However, they rely on committing a transaction proof to Ethereum to achieve finality, which can take hours to days.
In our benchmarks we just considered the L2 E2E latency time, which is not the same as L1 E2E latency with finality. The actual rollup process from L2 to Ethereum can take hours, followed by a 7-day challenge period on Ethereum. This means achieving irrevocable finality for a transaction could take up to a week on these L2 chains.
The Future of Blockchain is Faster
Our breakthrough in achieving sub-second end-to-end latency on Aptos marks a significant milestone in the evolution of blockchain technology. With this level of speed and efficiency, the promise of blockchain applications can finally reach their full potential of onboarding the next billion users to Web3. Moreover, as we continue to innovate and push boundaries, it’s evident that we’re only at the beginning of this journey toward even faster blockchain solutions.
