Fast Concurrency Control with Thread Activity Management beyond Backoff

抄録

We attempted to improve the performance of Silo, a concurrency control protocol for in-memory database management system that does not perform efficiently under high-contention workloads. Adaptive backoff is known as an effective optimization method under high-contention workloads. As a result of analysis, we found that its efficacy lies in the elimination of unknown conflicting events rather than in the reduction of re-conflicting events already known, which has been considered so far. On the basis of this analysis, we propose a new method, the thread activity management method (TAM). We conducted experiments comparing Cicada, the state of the art concurrency control protocol, and Silo with our proposed method. The results demonstrated that the proposed method enabled Silo to significantly outperform Cicada. We found that cache misses are vitally relevant to performance.