Scalable Compact NUMA-aware Lock

In the era of multicore processors and Non-Uniform Memory Access (NUMA) architectures, the need for efficient synchronization mechanisms that are aware of the underlying hardware topology is critical. This paper presents a Scalable Compact NUMA-aware (SCNA) lock, an innovative NUMAaware locking mechanism designed to optimize performance in modern computing environments. The SCNA lock considers a single queue and performs a few atomic operations for efficient lock handovers. The SCNA lock organizes threads based on their NUMA node affinity. It intelligently adapts to contention levels and the specific characteristics of NUMA architectures, thereby reducing remote memory accesses and improving cache locality. It prioritizes access to local memory nodes, thus minimizing access latency and maximizing throughput. Our extensive evaluations across several concurrent data structures demonstrate that SCNA lock consistently outperforms traditional locking mechanisms and existing NUMA-aware alternatives. The SCNA lock also offers a space-efficient alternative for NUMA-aware locking, maintaining constant memory usage regardless of NUMA node count.