Understanding Sui’s object model and scaling capabilities
Sui utilizes an object-centric data model where assets are Move objects with unique identifiers, allowing for high parallel throughput. The network demonstrated significant scaling by reaching 6,000 TPS during a peak in inscription activity in December 2023.
Objects instead of accounts
Sui uses an object-centric data storage model instead of a global balance ledger. In Ethereum, a user has a single number for a balance. On Sui, every asset is a Move object with a unique 32-byte identifier and owner. This means a transaction involving an asset consumes the old object and produces a new one. If you want to send 5 SUI, the system splits your current Coin object into two. One object contains the 5 SUI for the recipient and the other contains the remaining balance for you. This fragmentation creates a UX problem because users face higher gas costs if they must merge many small coins to reach a target amount. I see that developers must use a greedy algorithm to pick the smallest number of coins to reach a target amount to reduce these costs. This process is atomic within a single TransactionBlock. Everything succeeds or everything rolls back. You should realize, since you already know how simple accounts work, that this complexity is a direct result of the Move resource model. Sui’s bytecode verifier ensures that the network always assigns fresh UIDs to new objects and never reuses identifiers. Transactions explicitly specify all inputs up front using unique identifiers to allow the network to schedule transactions with non-overlapping inputs in parallel.
Scaling and throughput
Sui scales by processing transactions in parallel. The network uses a state access method where transactions declare which objects they will touch. Because every object maintains an independent version history and carries its own unique metadata, the network can process transactions affecting different objects concurrently without the need to synchronize a global state across various shards or use complex inter-shard messaging. Sui reached 6,000 TPS during a peak in inscription activity in December 2023. A test of programmable offchain tunnels for AI agents reached 40.6 million TPS. The Pilotfish prototype demonstrates linear scaling. When backed by 8 machines, the system increased throughput by 8x. This setup keeps latency below 20ms. The time-to-finality is approximately 390 milliseconds. The consensus engine uses Bullshark and Narwhal algorithms to arrange data efficiently. Pilotfish specifically partitions the validation function into three roles: the Primary, SequencingWorkers, and ExecutionWorkers.
| Metric | Value |
|---|---|
| Peak TPS (Offchain Test) | 40.6 million |
| Peak TPS (Inscription Surge) | 6,000 |
| Time-to-finality | ~390 milliseconds |
| Total SUI Supply | 10 billion |
| SUI in Circulation | ~3 billion |
Security and the Move language
The Move language on Sui provides security through capability objects. Instead of checking a sender’s address, the system requires a capability object to prove permission. This architecture prevents reentrancy attacks because the type system ensures a resource cannot be reused once a transaction moves it. I find that the separation of code from data is the most significant feature. This design allows users to manage assets with higher precision than account-based chains. However, I find that switching to Sui requires unlearning assumptions from the Ethereum model. Users must manage their own coin fragmentation to avoid excessive fees. The SUI token has a total supply of 10 billion. Approximately 3 billion SUI are currently in circulation. Of the total supply, 50% went to the Community Reserve for research and validator subsidies. Another 14% went to investors and 10% went to the Mysten Labs treasury. This makes the SUI token a versatile asset for staking, fee payment, and governance. I wonder if the growing number of tiny coin objects will eventually force a change in how users interact with wallets.
Join the discussion