Distributed Transaction Performance and Consistency Enhancements Based on Percolator

With the rapid development of the Internet and computer technology, the volume of data and user scale in web applications has been experiencing explosive growth. Due to this massive data, the storage and transaction processing demands cannot be met by traditional relational databases. Therefore, the focus of this paper is on the research of distributed transactions based on the Percolator distributed transaction model in conjunction with TiDB. To address transactional performance, an in-memory lock table mechanism and a crash recovery strategy are proposed to reduce transaction processing latency. In terms of transactional consistency, the S-Percolator protocol is introduced, which is based on read-write conflict detection, ensuring the serializability of transaction scheduling sequences. Through experimental results, the effectiveness of these improvements in TiDB is demonstrated, with reduced database transaction processing latency and the achievement of a serializable isolation level of concurrency control.