Poplar: Partially-Ordered Parallel Logging for Lower Isolation Levels

Existing parallel logging schemes are unsuitable for lower isolation levels due to tracking all dependencies of transactions. To overcome this problem, we present a high-performance parallel logging scheme (Poplar) in database management systems (DBMSs), which is compatible with multiple concurrency control techniques. Poplar uses a partially-ordered log sequence number (PSN) to encode write/read-dependencies of transactions. These dependencies are sufficient to ensure the correctness of logging and recovery. Our experimental evaluation shows that Poplar outperforms the state-of-the-art parallel logging schemes by up to \(\sim 67\%\) at Read Committed and \(\sim 23\%\) at Repeatable Read in the YCSB workload, and achieves up to \(\sim 97\%\) better performance at Read Committed in the TPC-C workload. It also enables the DBMS to recover up to \(\sim 22\%\) faster than these baselines.