Distributed Computation and Large-Scale Visualization in Heterogeneous Compute Environments

摘要

High performance computing still typically disregards heterogeneous environments and focuses on homogeneous clusters. While providing advantages and easing development, this fails to address the recent change toward heterogeneous computing infrastructure. With DIANA we presented an abstraction layer for unified access to local compute hardware including different hardware APIs. In this paper we extend DIANA, enabling transparent access to remote hardware. In this context, we demonstrate how to use DIANA for distributed computation and large-scale visualization of the finite-time Lyapunov exponent field in heterogeneous environments.