Scalable Serializable Snapshot Isolation for Multicore Systems

Since 1990's, Snapshot Isolation (SI) has been widely studied, and it was successfully deployed in commercial and open-source database engines. Berenson et al. showed that data consistency can be violated under SI. Recently, a new class of Serializable SI algorithms (SSI) has been proposed to achieve serializable execution while still allowing concurrency between reads and updates. In a prior study, we analyzed the multicore scalability of SSI-based database engines, and found that the previous implementations of SSI do not scale well as core counts of underlying hardware increase, due to latch contention on internal data structures. This paper proposes a new approach to implement SSI, detecting the required read-write conflict conditions at runtime without needing to latch the transaction manager's internal structures. We show how execution paths for SSI inside a database engine can be run in parallel on behalf of concurrent transactions. We prototyped our SSI implementation based on MySQL-5.6.10. It scales well on Intel 4-way 32 core machine and outperforms the original implementation by up to 20X, while sustaining its high throughput over 500 clients.