计算机科学
块链
数据库事务
事务处理
计算机网络
质量(理念)
服务质量
分布式计算
钥匙(锁)
分布式事务
方案(数学)
计算机安全
集合(抽象数据类型)
作者
Shengjie Guan,Zhen Xiao
出处
期刊:
[Institute of Electrical and Electronics Engineers]
日期:2026-07-01
卷期号:: 315-326
标识
DOI:10.1109/icws72778.2026.00049
摘要
Blockchain is a critical infrastructure for Web3.0, and its throughput dictates QoS. High-performance consensus protocols have shifted the throughput bottleneck from the consensus layer to the execution layer. While parallel execution mitigates this issue, existing methods remain inadequate. Read-write set oblivious approaches suffer from cascading aborts and costly re-execution under high contention, and read-write set aware approaches are hindered by restrictive state declaration requirements, thread waste under concentrated conflicts, and performance degradation from coarse-grained group-level re-execution. To address these challenges, we propose Tiga, an adaptive twolevel parallel blockchain transaction execution framework that enables inter- and intra-component parallelism. Tiga introduces an incremental write set design and, building on this, incorporates T-Cache to cache read-write set derivation rules for transactions with static execution logic, which account for approximately 83% of transactions on Ethereum, thereby eliminating redundant execution overhead. Tiga further employs an adaptive grouping mechanism that quantifies serial span and gas volume, merging lightly loaded components to reallocate threads to bottleneck components. Finally, upon divergence detection, Tiga utilizes a Minimal Sub-Graph Regrouping (MSGR) algorithm based on topological ordering to narrow the transaction re-execution scope. Comprehensive experimental evaluations demonstrate that Tiga achieves throughput improvements of up to$3.2 \times$over BlockSTM and$15.2 \times$over serial execution, while maintaining superior performance even under high contention.
科研通智能强力驱动
Strongly Powered by AbleSci AI