Fine-grained locality-aware parallel scheme for anisotropic mesh adaptation

摘要

In this paper, we provide a fine-grained parallel scheme for anisotropic mesh adaptation on NUMA architectures. Data dependencies are expressed by a graph for each kernel, and concurrency is extracted through fine-grained graph coloring. Tasks are structured into bulk-synchronous steps to avoid data races and to aggregate shared-data accesses. To ensure performance prediction, time cost and load imbalance are theoretically characterized. The devised scheme was evaluated on a 4 NUMA node (2-socket) machine, and a mean efficiency of 70% was reached on 32 cores for 3 kernels out of 4. The impact of irregular degree distribution and data layout on scalability is highlighted.