So long, Solana? The Rise of Blockchain's Parallel Universes
Published: 03 October 2024
Written by @vik0nchain as part of @cybercapital’s ongoing research on the current state of parallelized scaling
As Solana's parallel execution model continues to thrive, a new wave of projects- such as Sei, Aptos, and Sui- are emerging with innovative approaches to further optimize the transaction process and blockchain state management. Parallelization solves the traditional bottleneck of gen 1 blockchains such as Bitcoin and Ethereum by processing transactions simultaneously, rather than sequentially.
The primary challenge with parallel processing lies in conflicting transactions. Conflicts arise when multiple transactions want to access the same part of the state, forcing one party to wait for the other. Parallelized chains operate under the principle of independence, in which transactions are considered independent if they do not access any shared portions of state. A chain's conflict resolution implementation serves as the key differentiator in the efficiency and effectiveness amongst the recent wave of parallelized chains.
When Solana’s testnet went live in 2020, it showcased its account-based Proof-of-History (PoH) as the premier solution. In account-based models, transactions explicitly specify all accounts they interact with. To do this, Solana leverages PoH, a cryptographic clock-like sequence, to establish a verifiable sequence of events on the network. Transactions are then forwarded to the current and upcoming block leaders via Gulfstream, Solana’s transaction management system, eliminating the need for a traditional mempool.
Leaders are predetermined through a leader schedule based on PoH, which allocates validators to produce blocks in specific 400-millisecond slots, assigned via a stake-weighted random selection. This schedule is established once per epoch, approximately every two days, ensuring continuous block production. Preprocessing enables validators to achieve “readiness” before block inclusion, significantly reducing latency and enhancing throughput. Once they reach validators, transactions are executed in parallel by Solana’s execution environment, Sealevel, which analyzes which accounts are included in a transaction’s read-write set to distribute tasks across multiple threads. During execution, conflicting accounts are temporarily locked until the transaction updates the state, allowing non-conflicting transactions to run concurrently, while ensuring conflicts are processed concurrently.
In traditional AMM’s, many users compete to access the same liquidity pools and tokens. The trade off in such systems is that the first transaction to execute takes priority, leaving subsequent transactions unprocessed if another user's action changed the state. This “lock up” creates an unfavorable user experience, as delays expose transactions to front running where opportunistic parties, as seen with MEV in Ethereum, can interact with the pool and change the price, causing their actions to take effect under conditions other than those originally intended. This can be particularly costly for trading algorithms.
While Solana’s innovations set a precedent for scalability, new competitors continue to innovate around the bottleneck of conflict management, to reduce or avoid account locking entirely. Sei took Solana’s parallelized model and combined it with an implementation of the Ethereum's Virtual Machine (EVM), with the additional innovation of optimistic concurrency. Built on the premise that conflicts are rare, optimistic concurrency increases throughput by processing transactions assuming there are no conflicts, effectively pushing the problem from developers to validators. If conflicts are detected, they are rolled back and retried concurrently. This avoids the lock-up bottleneck, optimizing throughput and minimizing latency, with the addition of EVM compatibility, at the cost of loss of welfare in the form of additional gas fee’s, diminished composability, and delay induced slippage when conflicts arise. In economics, a loss of welfare describes inefficiencies that reduce the net benefit to participants and yield negative externalities. In the case of efficient scalability, this is akin to the additional costs—whether in gas fees or slippage—when transaction rollbacks occur, ultimately reducing efficiency and user experience. It should be noted this design shifts the bottleneck from the account lock up, to the rate of conflict. As rollbacks of conflicting transactions occur post-execution, the new bottleneck is best described as the relationship between number of conflicts and amount of transactions. Specifically, the cost of resolving vs. the number of conflicts is constant. This doesn’t happen in Solana because it has variable rollback costs. While both Solana and Sei implement variations of optimistic concurrency, where transactions are initially processed under the assumption of no conflicts. However, Solana’s design leverages PoH to manage rollbacks pre-execution by dynamically adjusting the rollback overhead based on the transaction’s position in the pipeline. In contrast, Sei’s model processes rollbacks post-execution, which incurs higher computational and network costs as the transactions must be reprocessed after failure, introducing latency and resource strain not present in Solana’s real-time handling of conflicts.
Move-ing forward
While Sei's approach improves upon account lock-up, the emergence of Move language, an object oriented model with favorable design to ease conflict management, presents further room for innovation. Move represents interactions amongst objects- best understood as assets- rather than accounts. Owned objects can be modified or transferred without interacting with other parts of state. This allows the system to filter for potential conflicts at the asset level, categorizing them as either shared or owned objects, providing a more efficient and flexible foundation for parallel processing.
Beyond the scope of scalability, Move introduces distinct features as granular management of transaction costs for owned versus shared objects, gas rebates as an incentivize to free up block space, and Gas Buckets which help dis-incentivize developer hyper-optimization, offering a safer and more adaptable environment compared to Ethereum's Solidity.
The Move language was originally invented by Meta's blockchain division. When the project ran into regulatory issues, multiple team members left Meta, some of which started Aptos and Sui.
Aptos leverages Move’s object based orientation to further mitigate conflicts of optimistic concurrency in parallel execution. While the logic of owned vs shared objects significantly reduces the likelihood of conflicts by pre-classifying owned objects, which are non-conflicting by nature, they are still limited by the continuous need to monitor read and write to different parts of the state.
Unlike Sei, which rolls back and retries conflicts post-execution, Aptos's Software Transactional Memory mechanism “Block-STM” dynamically detects and resolves conflicts during execution, selectively rolling back only the conflicting transactions. This ensures non-conflicting transactions proceed uninterrupted, reducing the penalty for incorrect transactions while optimizing throughput and minimizing latency. Thus, Aptos shifts the bottleneck from the rate of conflict to the computational overhead of monitoring read and writes across state, which bears a linear to sublinear relationship to the complexity and number of objects being processed concurrently.
The greater issue amongst optimistic concurrency in parallelized chains over the long term, is state bloat. Currently, about 95% of ETH transactions interact with smart contracts, of which 75% interact with a minimum of 3 smart contracts independently. This is particularly concerning for DeFi, who’s volume heavily relies on composability. As complexity and size of state grows, the likelihood of conflicts inevitably rises in tandem, while the cost of resolving them remains constant, creating a linear to sublinear curve to the bottleneck over time.
While Aptos and Sei both leverage optimistic concurrency to improve scalability, Sui opts for a deterministic approach, pre-ordering non-conflicting transactions prior to- rather than during or post-consensus, effectively eliminating conflict related rollbacks. To do this, Sui leverages Narwhal, a Directed Acyclic Graph (DAG) structure which serves as a mempool and data availability layer. By preserving the conflict-avoidant classification dichotomy of Move, Narwhal is able to achieve both deterministic and efficient pre-ordering transactions, avoiding conflicts in execution.
In traditional models like Bitcoin and Ethereum, duplicate broadcasting of transactions- first to each validator’s mempool, then to the block proposer for inclusion in the next block- creates redundant communication and congests the network. Narwhal drastically reduces latency by allowing validators to independently parse their local DAG structure to sort transactions while minimizing communication overhead.
Once ordered, Bullshark, Sui’s consensus protocol, finalizes transactions via Byzantine Fault Tolerant (BFT). Unlike traditional consensus models, Bullshark operates with partial synchronicity, where the network functions asynchronously until it reaches a point of Global Stable Time (GST), at which consensus can be carried out deterministically with synchronous communication. This design minimizes the impact of delays, temporary partitions, and other variable conditions without sacrificing time to finality.
The primary bottleneck of Sui lies in the linear relationship between latency and sync as the number of validators grows, yielding a more predictable scaling curve than its competitors, for which latency and sync are a secondary bottleneck relative to managing the ever-growing complexity of state with an optimistic model. To avoid network degradation, each of these parallelized protocols has implemented strategies to discreetly disincentivize new validators, such as substantial staking-capital requirements and capping voting power of individual validators. Nonetheless, joining the network is still permissionless, and any party who meets these steep requirements can participate in the network.
Unlike rollbacks, latency doesn’t bear a direct cost of resolution, as seen with Sei or Aptos, eliminating the loss of welfare innate to optimistic architecture. Furthermore, latency can be improved through dynamic management of resources, as seen by the unveiling of Aptos’s Shaol++ and the main net launch of Sui’s Mysticeti which shortly followed. Sui’s Mysticeti launch in May 2024 reduced network latency by up to 80% through dynamic quorum selection, minimizing unnecessary messages by enabling subsets of validators to finalize transactions based on current network conditions.
Models and Curves
While fruitful, the long-term implications of such designs are of greater interest, best understood through bottleneck curves, which portray the primary bottleneck's development when a protocol is scaled horizontally through additional validators.Improvements result in one of two outcomes, either movement along, or shifting a given bottleneck curve.
Movements along the bottleneck curve are a result of incremental changes which improve performance within the existing architecture, such as reducing conflict rates or enhancing validator coordination. While these improvements can lead to better efficiency in the short term, they fail to address the root causes of inefficiencies that lead to a loss of welfare. In this context, loss of welfare manifests as increased transaction costs, higher latency, and diminished user experience due to inefficiencies like transaction rollbacks and resource wastage.
In contrast, shifting the bottleneck curve involves structural changes, such as redesigning conflict detection mechanisms or altering the way state is managed. These changes lower the inherent inefficiencies of the system, reducing the intensity of the bottleneck, albeit with the common tradeoff of introducing new layers of complexity. By fundamentally altering the protocol's architecture, shifting the bottleneck curve can significantly mitigate the loss of welfare experienced by network participants.
A flatter bottleneck curve implies that improvements yield more substantial gains in efficiency, directly benefiting users through lower transaction costs and enhanced network performance. This not only addresses immediate technical limitations but also reduces negative externalities—such as network congestion and inflated gas fees—that impact all users. Thus, structural changes which shift the bottleneck curve are quintessential for maintaining competitive advantage by efficiently showcasing sustainable scalability while preserving the economic incentives that drive network adoption and growth.
Parsing the Scalability Puzzle
With optimistic models, the sublinear relationship of conflict resolution costs is bound to cause a loss of welfare, which grows in sublinear relation to state, as the growth rate of interdependencies exceeds the rate of state growth. State growth refers to the increasing complexity of the ledger over time, as more transactions and interdependencies accumulate. This puts both a strain on computational resources as well as higher chances of transaction conflicts. Over time, this can slow down network performance and affect scalability, as validators are required to manage larger states efficiently. This makes optimistic concurrency with an account-based language especially challenging, as updates return marginal, rather than linear improvement. Thanks to real-time conflict detection enabled by Move allowing non-conflicting transactions to be run without re-execution, Aptos effectively reduces said loss of welfare, and faces a less steep, but still sublinear, bottleneck due to growing complexity of overhead being managed by the STM. This makes Sui’s design particularly attractive, as its flatter bottleneck curve implies future improvements will yield to linears improvements.
Sei, Aptos, and Sui Relative Bottleneck Curves
Looking forward, it will be quintessential for Sei to prioritize shifting its bottleneck curve to improve composability with scale and remain competitive in DeFi to secure its place as the fastest scalable EVM chain. Aptos will be implementing Dynamic Dispatch in Q4 2024, optimizing runtime function flexibility by allowing contracts to determine which functions they execute during operation rather than prior, shifting its bottleneck curve to a flatter position. This change will significantly differentiate their implementation of Move from that of Sui, improving functionality, while bringing new potential vulnerabilities for re-entrancy attacks. It's worth noting that Sui’s use of DAG architecture and Sei using optimistic solutions are also more likely to benefit from innovations of other projects which could be implemented, while Aptos’s STM is a bit more unique. As Sui’s deterministic execution and conflict-free design continues to lead, its next challenge of optimizing network synchronization and managing validator growth will be critical to maintain its competitive edge in performance as the ecosystem expands.
In the past, Solana took a great leap in this direction by leveraging transactions for message passing, effectively reducing latency and sync to transaction speed. We at Cyber Capital believe a similar approach could alleviate this bottleneck, eliminating the same validator soft-cap while simultaneously improving both scalability and decentralization. As these chains are already hyper scalable, the additional network stress caused by communication transactions presents a viable and proven solution. While Solana continues to press the lead with its latest Frankendancer reveal, projects must navigate the delicate balance of validator expansion and state bloat while maintaining security and composability to capture the demand of users.
For more research and analytics on hyper scalable monolithic chains, follow me at @vik0nchain, Cyber Capital @cybercapital, or our founder @Justin_Bons on X.
Looking for a deeper dive into these parallel universes? Check out the resources below, or, take advantage of low fees and try these chains yourself!
Key resources: Life of a Solana Transaction, Optimistic concurrency, Aptos’ Block STM, Mysticeti Update Sui, Aptos’ Shaol++
