May 2014 Summaries

Rescale simplifies the process of running simulations against various configurations by offering a design of experiments (DOE) approach, also known as a "Parallel Process" within its interface. The setup begins with an existing job that includes selected hardware, software, input files, and analysis commands, which are then modified to create a DOE by parameterizing input files into Rescale templates and specifying unique combinations of variables for each "run." Users can define these combinations in a browser or through a CSV file, allowing for a dense distribution in the design space, and can increase the number of cores to enable parallel execution of runs. The Rescale platform handles the execution by replacing placeholders in the template with the specified values, generating results that can be reviewed using its plotting tools. This process ensures efficient execution of multiple simulations concurrently, enhancing computational capabilities and reducing the time required for analysis.

Microsoft's Big Compute solution introduces InfiniBand connectivity with RDMA support for its new A8 and A9 cloud instances, addressing the challenge of slow interconnect speeds often found in cloud-based high-performance computing (HPC) compared to on-premise clusters. This development is particularly significant as it offers near bare-metal performance, making it attractive for enterprises needing high-bandwidth, low-latency networks for tightly coupled simulations. However, the solution's current limitations include the requirement for applications to run on Windows and the complexity of configuring an MPI cluster on Windows compared to Linux, which most HPC practitioners are accustomed to. While Microsoft provides tools like the HPC Pack to assist with setup, its perceived complexity and the need for applications to be recompiled against MS-MPI libraries may pose barriers to entry. Despite these challenges, the impressive performance metrics achieved in latency and bandwidth tests suggest the potential of Big Compute, especially as future support for Linux VMs could further enhance its appeal in the HPC market.

Rescale is enhancing its platform by developing a public-facing API to allow programmatic job bursting, a popular customer request. The platform, initially built on a Django application, is being transitioned to a RESTful API using the Django REST Framework (DRF), which has streamlined the code by making custom validation redundant. A Rescale job is defined by job analyses and job variables, where job variables serve as input parameters and job analyses specify the simulation software along with necessary commands and hardware configurations. The new API allows users to upload files, construct job JSON data, and manage jobs through endpoints, and although currently in private beta, it offers a Python client with limited API calls like file upload, job creation, and status monitoring. Interested users can contact Rescale for access to the API, which promises to elevate their computational capabilities.

Siemens’ NX Nastran software is now accessible through Rescale’s cloud-based, dynamically scalable platform, enabling users to perform large-scale simulations efficiently and cost-effectively with a pay-per-use model. This integration allows engineers to customize compute capacity for virtual product simulations, enhancing efficiency, product quality, and speeding up time-to-market. The platform supports executing multiple simulations simultaneously, including Designs of Experiment (DoE), which helps in understanding parameter variations across design spaces. This collaboration provides Siemens PLM Software customers with flexible options to augment their existing high-performance computing capabilities without investing in additional IT infrastructure. Rescale's system, which partners with leading software vendors, facilitates instant access to various simulation packages, offering customizable HPC hardware for engineering and scientific simulations.