Design, Analysis, Implementation and Deployment of a High-Performance, Out-of-Core, Parallel, Dense Direct Linear Solver

摘要

We present the design, implementation and deployment of SPIDO2, a massively parallel out-Of-Core direct dense linear solver used by Boundary Element Method (BEM) codes developed at EADS Innovation Works, and currently in production in all EADS business units. It is designed as a high performance drop-in replacement for a legacy solver, uniformly achieves very good scalability up to several hundreds cores (>80% of peak performance), is portable and takes into account the new context of HPC, leveraging OpenMP and MPI for intra/inter-node parallelism. We provide here a detailed analysis of the algorithms and data structures used to achieve a very efficient out-of-core parallel decomposition (LU and LDL~t), including run-time complexity validating the design choices. We then describe the implementation, benchmarks and validation methodology that led to the deployment of this solver in the BEM codes, and we finally provide insights on how we expect this design to be well-suited to the current and future advances in HPC technology.