An Asynchronous Runtime System using RDMA and Lightweight Tasks
|Title||An Asynchronous Runtime System using RDMA and Lightweight Tasks|
|Year of Publication||2015|
|Authors||Dang, H, Brooks, A, Dryden, N, Snir, M|
We describe a novel runtime system that integrates a task model with RDMA communication and software caching. For evaluating the runtime system, we design two microbench marks and implement three applications: Barnes-Hut, Sparse triangular linear solver, and Monte Carlo particle tracking. The resulting codes are simpler, since load balancing across cores and latency hiding are hidden in the runtime. This also results in them performing better than state-of-the-art implementations of the same algorithms by up to 13, 10.8, and 3 times, respectively.