| 薛晗,包云岗.面向EOV联盟链的高效隔离性校验方法[J].高技术通讯(中文),2024,34(11):1142~1152 |
| 面向EOV联盟链的高效隔离性校验方法 |
| High-efficiency isolation verification methods for EOV consortium blockchain |
| |
| DOI： |
| 中文关键词: 区块链； 多版本并发控制（MVCC）； 缓存； 并行化 |
| 英文关键词: blockchain, multi-version concurrency control（MVCC）, cache, parallelization |
| 基金项目: |
| 作者 | 单位 | | 薛晗 | （中国科学院计算技术研究所北京 100190）
（中国科学院大学北京 100049） | | 包云岗 | |
|
| 摘要点击次数: 17 |
| 全文下载次数: 21 |
| 中文摘要: |
| Hyperledger Fabric等采用交易预执行(EOV)模型的区块链系统，通常采用多版本并发控制（MVCC）方法校验交易，其基本思想是在提交前校验当前交易的读集合是否已经被跨块和块内的交易修改过，以保证交易隔离性和正确性，然而MVCC常因为并行化低和频繁底层数据库访问成为整个交易周期的性能瓶颈。为解决此问题，本文提出了2种新MVCC算法：C-MVCC和P-MVCC。C-MVCC以内存作为缓存，减少MVCC校验时对底层数据库系统的访问，并且针对跨片校验阶段引入并行化机制；P-MVCC则是在C-MVCC的基础上进一步拆解交易间的依赖关系，同时实现了跨块和块内校验的并行化，从而消除分片场景下过高的跨片校验时延累积。实验结果表明，C-MVCC在一般场景下相较于当前的算法可以带来7.5倍的校验效率提升，而P-MVCC在分片场景中可以带来9.0倍的校验效率提升。 |
| 英文摘要: |
| Blockchain systems such as Hyperledger Fabric that adopts the execute-order-validate (EOV) transaction model usually use the multi-version concurrency control (MVCC) method to validate transactions. The basic idea is to check whether the read set of the current transaction has been modified by cross-block and intra-block transactions before committing, to ensure transaction isolation and correctness. However, MVCC often becomes the performance bottleneck of the entire transaction due to low parallelization and frequent access to the underlying database system. To solve this problem, this paper proposes two new MVCC algorithms: C-MVCC and P-MVCC. C-MVCC uses memory as a cache to reduce access to the underlying database system during MVCC validation, and introduces a parallelization mechanism for the cross-shard validation stage. P-MVCC further decomposes the dependency relationship between transactions based on C-MVCC, and realizes parallelization of cross-block and intra-block validation, thus eliminating the accumulation of high cross-shard validation latency in sharding scenarios. Experimental results show that C-MVCC can bring 7.5 times improvement in validation efficiency compared with the current algorithm in general scenarios, while P-MVCC can bring 9.0 times improvement in validation efficiency in sharding scenarios. |
|
查看全文
查看/发表评论 下载PDF阅读器 |
| 关闭 |
