Aptos
Benchmark de Desempenho Block-STM
Context
This figure appears in the Protocol Design section of the Aptos whitepaper, specifically within the discussion of the Block-STM parallel execution engine. It follows the technical description of Block-STM's multi-version data structure and optimistic concurrency control algorithm, and is presented as empirical validation of those design claims. The benchmark was conducted as an isolated, execution-only experiment using non-trivial peer-to-peer Move transactions with an in-memory database.
What This Figure Shows
The chart plots transaction throughput against the number of physical CPU cores under two contention conditions: low contention (many accounts, few conflicts) and high contention (few accounts, frequent conflicts). Under low contention, Block-STM achieves a 16x speedup over sequential execution when scaled to 32 threads, demonstrating near-linear scaling with core count. Under high contention, the speedup remains above 8x at 32 threads, showing that the engine degrades gracefully rather than catastrophically when transactions conflict. Block-STM uses a multi-version data structure where all writes to the same memory location are stored with versioned transaction IDs; readers obtain the value written by the highest-numbered preceding transaction. Unsuccessful post-execution validations trigger re-execution of affected transactions, and the engine uses delta writes to further reduce conflict surface on shared counters and accumulators.
Significance
This benchmark is the primary empirical justification for Aptos's claim to high-throughput smart contract execution, distinguishing it architecturally from single-threaded EVM chains and from parallel environments that require programmer annotation. It demonstrates that Block-STM's adaptive, hint-free approach can capture meaningful concurrency even under adverse contention conditions, which is the realistic operating environment for a popular public blockchain.