Exascale_Computing_without_Threads-Barry_Smith.pdf

<div> <div> <div> <p>We provide technical details on why we feel the use of threads does not offer any fundamental performance advantage over using processes for high-performance computing and hence why we plan to extend PETSc to exascale (on emerging architectures) using node-aware MPI techniques, including neighborhood collectives and portable shared memory within a node, instead of threads. </p> </div> </div> </div>