Deadlocks: A Critical Performance Bottleneck in High-Contention OLTP Systems

High-contention Online Transaction Processing (OLTP) workloads pose a critical challenge for concurrency control protocols, particularly due to hotspots leading to excessive transaction aborts and deadlocks. Traditional approaches, including Two-Phase Locking (2PL) [1], struggle under such conditions, suffering from performance degradation due to wasted work from aborts and the harmful effects of transaction retrials. This proposal introduces B2PL, a novel deadlockfree 2PL protocol designed to tolerate high contention. B2PL leverages static analysis of transactions using a new graph-based structure, SLWGraph, to predetermine lock acquisitions and enforce an ordering that eliminates deadlocks within the 2PL framework. Furthermore, it enhances parallelism by enabling early lock release through a flexible mechanism called partial transaction chopping. By avoiding deadlocks and subsequent retries, B2PL aims to significantly improve throughput and latency in demanding OLTP environments. Our evaluations show promising results, indicating substantial performance gains over state-of-the-art 2PL protocols [2].