Novel HPC Technologies for Scalable CAE: The Case for Parallel I/O and File Systems

As HPC continues its aggressive platform migration from proprietary supercomputers and Unix servers to HPC clusters, expectations grow for clusters to meet the I/O demands of increasing fidelity in CAE modeling and data management in the CAE workflow. Cluster deployments have increased as organizations seek ways to costeffectively grow compute resources for CAE applications, and during this migration many also implemented conventional network attached storage (NAS) architectures to simplify IT administration and further reduce costs. While legacy NAS implementations offer several advantages of shared file systems, most are too limited in scalability for effective management of I/O demands with parallel CAE applications. As such, a new storage migration is underway to replace legacy (serial) NAS with parallel NAS architectures and parallel file systems. This new class of parallel file system and shared storage technology was developed to scale I/O in order to extend the overall scalability of CAE simulations on clusters. This paper examines CAE motivation for shared parallel file systems and storage, for requirements of multi-physics LS-DYNA applications on conventional clusters with proper balance for I/O. Model parameters such as size, element types, schemes of implicit and explicit (and coupled), and a variety of simulation conditions can produce a wide range of computational behavior and I/O data management demands. The benefits of a Panasas storage implementation are introduced for such broad requirements, through examples of CAE workflows for a variety of production-level applications in industry.