Optimizing Aria Concurrency Control Protocol with Early Dependency Resolution

Aria is a deterministic concurrency control protocol for long-running or highly contended transaction processing workloads. While Aria provides excellent performance for its concurrent execution of conflicting transactions in the execution phase, the feature can provide inefficiency in some cases. To address the issue, this paper proposes a novel optimization method, early dependency resolution. The proposed method divides the dependency checking of Aria into two stages. The first stage checks only the write-after-write dependency, while the second stage checks the remaining read-after-write and write-after-read dependencies. Early aborting in the first stage allows the second stage to reduce aborts due to false positives and checking overhead. Consequently, the proposed method expands the scheduling space from vanilla Aria and improves performance. We implemented Aria and the proposed method based on CCBench system and compare it with vanilla Aria in experiments. The results show that the proposed method outperforms Aria in high-contention and long-running workloads, with a performance difference of up to 36 times.