NUMA aware locks Implementation and Evaluation

Programs running on NUMA machines are sensitive to memory access locality[3]. Acessing data on the local memory is significantly faster than remote memory. Thus, designing locks that can take advantage of this property would improve the performance. We review three papers related to this topic and study the behaviors of these NUMA aware locks. We implement these locks in C++, and evaluate them on the two different NUMA platforms–intel and sparc. We compared these locks each other and with non NUMA aware locks in terms of performance and fairness.