In May 2010, we started from the ground up to design and develop OceanBase [6, 7], a distributed relational database system based on commodity hardware. OceanBase was first collected by Taobao in 2011 [3], a service similar to Amazon.com's wish list [11]. It has since been used by Alipay in 2014, and by Zhejiang E-Commerce Bank in 2015, as well as many other commercial banks, insurance companies, and other organizations for communication and energy applications.
This paper first introduces OceanBase's detailed design goals and standards, system architecture, SQL engine, and multi-tenancy in §2. Second, it proposes an LSM tree-based [35] storage engine, and discusses asymmetric read-write design, daily incremental primary compaction, and replica types in Section 3. Third, in §4, it proposes a transaction processing engine, including timestamp services, transaction processing, isolation levels, and replicated tables in OceanBase. Fourth, in Section 5 we conduct the 2020 TPC-C benchmark on OceanBase. Section 6 presents the lessons learned in building OceanBase. §7 briefly reviews related work. Finally, we conclude our work in Section 8. We briefly list our contributions in the following items.
• We have built OceanBase since 2010, a distributed relational database. As a horizontally scalable multi-tenant system, OceanBase has cross-region fault tolerance and supports a shared-nothing architecture. When a few nodes fail, its RPO (Recovery Point Objective) becomes zero and RTO (Recovery Time Objective) is less than 30 seconds.
• We proposed a oltp database engine based on the LSM tree, which has been optimized many times, and its performance is close to that of an in-memory database. Designed and implemented an asymmetric read-write data block storage system and daily incremental Major Compaction.
• We propose a Paxos-based 2PC, named OceanBase 2PC, to improve distributed transaction processing capabilities and reduce transaction delays. It introduces the Paxos protocol into 2PC, so that distributed transactions have automatic fault tolerance. Compared with traditional 2PC, the state of the coordinator in OceanBase 2PC is not persisted, thus reducing the number of Paxos synchronizations from three to two, and further truncating
Transaction latency of only one Paxos sync. • We conducted OceanBase's TPC-C benchmark test, reaching 707 million tpmC in 2020, the best in the world
record, so far.