Distributed array data management on NUMA multiprocessors

Management of program data to reduce false sharing and improve locality is critical for scaling performance on NUMA multiprocessors. We use HPF-like directives to partition and place arrays in data-parallel applications on Hector, a shared-memory NUMA multiprocessor. We present experimental results that demonstrate the magnitude of the performance improvement attainable when our proposed array management schemes are used instead of the operating system management policies. We then describe a compiler system for automating the partitioning and placement of arrays. The compiler uses a number of techniques that exploit Hector's shared memory architecture to efficiently implement distributed arrays. Experimental results from a prototype implementation demonstrate the effectiveness of these techniques.<
>