The state tree used for blockchain state management has emerged as a core performance bottleneck, primarily due to the latency involved in generating cryptographic commitments during block production and validation. To address this issue, we present a delayed write mechanism that decouples commitment computation from block execution. We introduce the Sliding Window Merkle Tree (SWMT), a cyclic in-memory cache that temporarily buffers recent state updates, deferring their integration into the state tree. SWMT uses a 32-bit window field to efficiently manage cache lifespan and support rapid commitment generation. To handle fluctuating transaction loads, we propose a TCP-inspired congestion control strategy that adapts caching granularity to dynamic workload intensity. Our design can be integrated with existing MPT and Verkle tree structures, while preserving proof succinctness and verifiability. Evaluation on 6 million real Ethereum blocks demonstrates up to a 61.3x reduction in commitment latency, a 6.3x speedup in block execution. These results highlight a practical and deployable approach to enhancing blockchain throughput and scalability.
Sheng et al. (Thu,) studied this question.