The Sui Smart Contracts Platform

Authors Blackshear, Sam; Cheng, Evan; Clark, Isaiah; Dill, David L.; et al. (Mysten Labs)
Year 2022
Project Sui
License Apache 2.0
Official Source https://github.com/MystenLabs/sui/blob/main/doc/paper/sui.pdf

This page is an educational summary and analysis of an official whitepaper or technical paper, written for reference purposes. It is not a verbatim reproduction. CryptoGloss does not claim authorship of the original work. All intellectual property rights remain with the original author(s). The official document is linked above.

“The Sui Smart Contracts Platform” is the 2022 technical paper by Mysten Labs describing the architecture of the Sui Layer 1 blockchain. Mysten Labs was founded by former Meta/Diem engineers — including Sam Blackshear (creator of the Move language), Evan Cheng, and George Danezis — who forked the Diem project’s Move language and re-architected the execution model around a fundamentally different primitive: objects rather than accounts.

The paper’s two main contributions are:

  1. A novel object-centric execution model that makes transaction parallelism explicit and first-class
  2. The Narwhal + Bullshark DAG-based consensus/ordering protocol that separates data dissemination from ordering for high throughput

> Paper: Available at github.com/MystenLabs/sui.


Publication and Context

Like Aptos, Sui is a product of the Diem collapse. But Mysten Labs made more radical architectural departures from the Diem design — particularly rejecting the account model entirely in favor of objects, and adopting a DAG-based mempool (Narwhal) for data dissemination.

Sui launched mainnet in May 2023. Its object model and consensus design have been the subject of academic discussion for their implications on parallel execution and cross-shard coordination.


Object-Centric Execution Model

In Ethereum and most blockchains, state is organized by account: each address has a balance and storage. In Sui, the fundamental unit of state is an object:

  • Every object has a globally unique ID, a type, and an owner
  • Objects are either owned (by an address or another object) or shared (accessible globally)
  • Smart contracts operate on objects — functions take objects as inputs and produce new or modified objects as outputs

Consequence for parallelism:

  • Transactions that only touch owned objects require no consensus — they need only a quorum of validators to confirm the sender’s authorization (a type of Byzantine broadcast). This is called the fast path and can achieve sub-second finality.
  • Transactions that touch shared objects require full consensus ordering (Bullshark) to prevent conflicting reads/writes.

This explicit distinction between owned-object transactions and shared-object transactions allows Sui to achieve natural parallelism without STM re-execution: non-overlapping transactions simply never conflict.


Sui Move Language

Sui uses a variant of Move (the Diem-origin language) with Sui-specific modifications:

  • Object-based storage: Move modules operate on Sui objects; storage is object-centric rather than account-map-centric
  • Transfer patterns: Objects can be transferred, shared, frozen (made immutable), or wrapped inside other objects
  • Capabilities: Sui Move uses capability-based access control — possessing a capability object grants permission to call certain functions

Sui Move diverged from Aptos Move in several specifics of the object model and standard library, creating two incompatible dialects of Move despite shared origins.


Narwhal: High-Throughput Mempool

Narwhal (Spiegelman et al., 2022) is a DAG-structured mempool designed to separate data dissemination from ordering:

  • Each validator continuously proposes blocks of transactions and broadcasts them to other validators
  • Validators sign off on blocks they’ve received, creating certificates (quorum-signed block receipts)
  • These certificates form a DAG where each block explicitly references prior blocks from other validators

The key insight: by the time an ordering protocol (Bullshark) runs, all referenced transaction data has already been disseminated and confirmed available. This eliminates the latency spike where consensus must wait for data propagation.

Bullshark is the consensus protocol that interprets the Narwhal DAG as a totally ordered sequence of transactions. It identifies anchors (designated blocks per validator per round) and orders the DAG by following anchor chains — a leaderless, high-throughput BFT ordering with formal safety proofs.


Consensus Properties

Property Narwhal+Bullshark
Safety BFT (tolerates f < n/3 Byzantine)
Finality ~2 rounds (sub-2 second typical)
Throughput > 100,000 TPS in lab benchmarks
Leader dependency Low (rotating anchor per round)
Message complexity O(n²) bandwidth for Narwhal

Reality Check

Sui’s object model is a genuine architectural innovation that makes parallelism declarative rather than speculative. In production, Sui has achieved meaningfully higher throughput than most EVM chains. However:

  • Most real DeFi transactions touch shared objects (AMM pools, lending protocols) which require full consensus and destroy the fast-path advantage.
  • Developer UX: The object model requires a significant mental shift from Solidity/EVM development. Most developers find the object-centric model harder to reason about for complex state patterns.
  • Move fragmentation: Sui Move and Aptos Move are incompatible, fragmenting the Move ecosystem.
  • Token distribution: Like Aptos, Sui’s initial SUI distribution was criticized for insider-heavy allocation.
  • Narwhal network costs: O(n²) bandwidth per Narwhal round limits practical validator set size.

Legacy

Narwhal and Bullshark have been published as standalone academic papers and have influenced the design of DAG-based BFT consensus beyond Sui — including Shoal (an improved Bullshark variant), Mysticeti (Sui’s next-generation consensus), and academic interest in DAG mempools as a throughput-scaling technique. The object model has driven interest in alternative state organization approaches beyond account-based models.


Related Terms


Research

  • Blackshear, S., Cheng, E., Clark, I., Dill, D., et al. (2022). The Sui Smart Contracts Platform. Mysten Labs.

— Primary paper. Section 3 defines the object model; Section 4 covers consensus; Section 5 the fast-path owned-object protocol.

  • Spiegelman, A., Giridharan, N., Sonnino, A., & Kokoris-Kogias, L. (2022). Bullshark: DAG BFT Protocols Made Practical. ACM CCS 2022.

— Formal treatment of Bullshark; proves safety and liveness under partial synchrony with Byzantine validators.

  • Danezis, G., Kokoris-Kogias, L., Sonnino, A., & Spiegelman, A. (2022). Narwhal and Tusk: A DAG-based Mempool and Efficient BFT Consensus. EuroSys 2022.

— Original Narwhal paper; demonstrates that separating dissemination from ordering allows 130,000 TPS on 50 validators.