Sei executes transactions in parallel and resolves the collisions afterward

How Sei runs transactions concurrently

Most blockchains process transactions one at a time. @SeiNetwork takes a different approach: it runs them all at once and sorts out any collisions afterward.

The engine behind this is Optimistic Concurrency Control (OCC). According to Sei's official documentation, OCC allows transactions to execute in parallel based on an initial estimate of the state each one will touch, rather than locking state resources upfront before execution begins. Multiple CPU cores handle non-conflicting transactions simultaneously, with the engine maintaining deterministic outcomes identical to sequential processing.

A conflict arises in one of two ways: when two transactions write to the same state key, or when one transaction reads state that another concurrent transaction has already written. The system identifies these clashes by comparing the actual read and write sets recorded during execution.

Conflict resolution and practical limits

When a conflict is detected, the affected transactions are re-run. A fixed resolution order prevents the process from looping indefinitely, and the final state always matches what sequential processing would have produced. As Sei's own research notes, OCC assumes conflicts are unlikely to be frequent, so the strategy pays off most when transactions are spread across different parts of state. Blocks heavily concentrated on a single contract see the smallest gains, since the collision rate climbs and more re-execution is required.

The parallelization engine applies to both EVM and CosmWasm environments on the network. Sei v2 introduced optimistic parallelization to remove the need for developers to pre-declare which state their transactions would access, simplifying the developer experience compared to earlier approaches used by chains like Solana and Sui, which require explicit dependency mappings before execution.

Sei's GitHub repository describes the chain as supporting "optimistic parallel execution of both EVM and CosmWasm," with Twin Turbo Consensus targeting 400ms block times alongside the parallelization gains.

Sources:
Sei Parallelization Engine, Sei Official Docs
Sei v2: The First Parallelized EVM, Sei Labs Blog
64.85% of Ethereum Transactions Can Be Parallelized, Sei Labs Research