Abstract

Engineers have long suffered from inadequate systems and software. Industrial surveys show that about 60% of engineers feel that hardware resources are inadequate for their work. Often, they have to simplify models to fit in current hardware limits thus losing accuracy of their results. In addition, buying HPC system comes with even more challenges, such as long procurement cycles; high Total Cost of Ownership (TCO); non-scalable software; erratic system usage; fast aging of inhouse servers; and the need for trained HPC system experts. But with the rise of cloud computing, it's easier than ever to get access to powerful computing, efficient storage, and low-latency networking. More and more organizations are turning to HPC cloud, due to additional business value coming from increased flexibility and scalability from practically unlimited, on-demand computing and storage capacity that can be provisioned if needed. The productivity of the simulation team grows thanks to the ability to run more projects at the same time or running more complex simulations, without big investments. Compute-intensive tasks such as parametric sweeps, multiphysics simulations, and large-scale optimization to evaluate many design options become possible. The other benefits include the possibility to increase the accuracy of their models by removing memory and CPU limitations so engineers can use more detailed models, access heterogeneous architectures such as GPUs, InfiniBand, large memory, latest processors, etc., and, finally, reducing engineering and licensing expenses by shorter computation time. However, running complex engineering workloads in the cloud is not easy. The engineering team needs cloud-specific know-how, the integration with existing on-premises infrastructure to ensure security and compliance, hands-on experience with high performant architectures, on-boarding complex applications to the cluster, optimizing software licensing cost for the cloud, and efficiently transfer large datasets. All components of compute, storage, and visualization should be optimally integrated with on-premises infrastructure into a single system. The UberCloud Platform is addressing these challenges by providing engineers a turn-key scalable engineering platform with automatically integrated compute, storage, and networking, optimized for specific CAE software, enabling engineers to significantly reduce IT overhead and modernize their IT infrastructure. The platform provides access to the latest hardware optimized for specific applications, high-resolution engineering virtual workstations for pre-post-processing and interactive simulation, scalable HPC, and automated update process without downtime, with minimal IT overhead, and the possibility to autoscale the cluster with tailored hardware for each specific job. We have demonstrated the feasibility of this approach at Freudenberg Technology and Innovation (FTI), an engineering facility providing expertise and simulation services to the Freudenberg Group, a global technology manufacturer with production sites in Europe, Asia, Australia, North and South America, and more than 48,000 employees. FTI faced the challenges of continuously updating the existing simulation environment to hardware for best performance and providing tailored hardware to their engineers for each specific simulation workflow. As part of their digital transformation, Freudenberg decided to migrate their engineering workloads step-by-step to Azure to provide engineers their customized simulation workflows on Azure, including a cloud-based engineering workstation. By moving their workflows to the UberCloud Platform on Azure, seamless hardware update, best performance, and tailored hardware configurations for each specific simulation task have been provided. In our presentation, we will explain the technical details of the migration process such as the cloud templates, the automated pipeline for the deployment, the simulation workflow containerization technology, and the software management. We will also cover the integration aspects of the UberCloud Engineering Simulation Platform, the CAE application containers with Azure CycleCloud, NetApp storage, and visualization. Last but not least, we will present benchmark results of key CAE applications running fully in the cloud on the latest AMD architectures.