Sui’s Mysticeti upgrade establishes the network as a speed leader

The Mysticeti consensus upgrade provides Sui with much higher efficiency and establishes the network as a speed leader. This uncertified DAG-based protocol reduces latency from 1900ms to 400ms. For owned object transactions, the fast path achieves finality in approximately 250ms. Mysticeti v2 integrates transaction validation directly into the consensus process to remove the computational overhead of redundant steps. This upgrade also introduces the Transaction Driver, which replaces the older Quorum Driver to lower CPU and bandwidth demands. I observe that this implementation reduces CPU usage for consensus by 40% in production environments. In Asia, nodes saw a 35% latency reduction, while Europe-based nodes saw a 25% reduction.

Mysticeti uses a novel commit rule that appoints multiple leaders every round to commit transactions earlier. The direct decision rule uses the DAG structure to provide a "virtual" certificate by checking if a block has enough witnesses and confirmations in subsequent rounds. This requires only two rounds of voting, which reduces round trips by 50% compared to the Narwhal-Tusk system. If the direct rule fails, the protocol uses an indirect decision rule to find an anchor in the DAG. This process helps avoid head-of-line blocking during pipelined rounds. I find the 40% CPU reduction significant, though the post-consensus checkpoint mechanism introduces latency for owned object transactions.

Sui’s object-centric architecture competes with Aptos

Sui’s object-centric architecture competes with the Block-STM parallel execution engine used by Aptos. Aptos claims throughput exceeding 160,000 TPS in benchmarks. Sui reaches 297,000 TPS in controlled testing environments. While Aptos uses Block-STM to reorder transactions and resolve conflicts during execution, Sui processes transactions in parallel by treating every asset as a unique object. You probably already know that both projects originated from Meta’s abandoned Diem project.

Aptos utilizes a modular design to separate consensus, data storage, and execution. This allows the Aptos team to implement upgrades with minimal downtime. Sui relies on its object model to allow validators to process non-overlapping transactions without a global state change. I prefer Sui’s ability to achieve 200,000 TPS with 0.5 second consensus commitment, especially since it avoids the complexity of sharding. Aptos achieves sub-second finality through its AptosBFTv4 protocol, which draws upon research from HotStuff and Jolteon. Aptos raised $350 million in 2022. Sui secured $336 million across its Series A and B rounds. Can the two networks ever bridge their technical differences?

The Move language heritage creates a fragmented development landscape

Sui developers use Sui Move, an adaptation of the Move language designed for objects, while Aptos developers work with the original Move implementation. This split creates separate ecosystems for smart contract development and developer tools.

Sui’s Move implementation includes specific ownership types: address-owned, immutable, party, shared, and wrapped objects. This allows for advanced patterns like transfer-to-object where one object controls another. This feature enables the development of smart contract wallets with access to objects based on dynamic policies. Aptos provides its own suite of tools, including the Aptos CLI and the Move Prover for formal verification. The divergence in how these chains handle resources stems from their different interpretations of the legacy left by the Diem project, and this forces developers to choose between two distinct paths to manage digital assets on-chain. Sui’s address-owned objects belong to an account address or an object ID. Aptos supports a wide range of SDKs in TypeScript, Python, Rust, and Go.