FTSD: a fissionable lock for multicores

Delegation is the highly efficient solution for parallel synchronization. However, the existing the-state-of-the-art delegation locks offer good performance at the cost of occupying computing cores under moderate contention, but exhibit sub-optimal single-thread performance and non-scalable performance under no and high contention, respectively. In this paper, we present a fissionable lock, called FTSD, which consists of two underlying locks: a TTS lock for lock stealing, which serves as a fast path, and a NUMA-aware delegation lock, which offers scalable performance under high contention. Our evaluation shows that FTSD delivers as good or better performance than other state-of-the-art locks.