March 2015 Summaries

Rescale provides various pricing plans for high-performance computing (HPC) simulations, including On Demand, Low Priority, and Prepaid options, with the Prepaid plan offering the lowest cost per hour in exchange for upfront payment. The decision to opt for a Prepaid plan depends on the utilization rate, with a break-even point at 27% utilization for the Nickel core type over a three-year period, leading to substantial savings if utilization is higher. The analysis involves determining the number of Prepaid cores to purchase based on a forecast of compute jobs, taking into account the total On Demand cost, Prepaid cost, and residual On Demand cost. By calculating residual core hours and performing programmatic analysis, users can optimize their hardware costs by maximizing Prepaid core utilization. The complexity of this calculation is managed through algorithms that account for overlapping intervals of compute jobs, and Rescale plans to introduce a Prepaid core calculator to assist users in making cost-effective decisions based on historical usage data.

Organizations are increasingly adopting hybrid and pure-play cloud-based High-Performance Computing (HPC) to maintain competitiveness, with a shift from on-premise clusters to scalable, secure, and elastic cloud environments. Joris Poort, CEO of Rescale Inc., highlights the industry shift towards simulation-driven design, which is enhancing product development through virtual prototyping and comprehensive simulations. This shift necessitates robust cloud HPC resources to handle variable demand, prompting companies to evaluate cloud providers based on global reach, hardware quality, and application availability. Rescale's cloud platform, tailored for Fortune 500 firms, offers a vast hardware footprint and extensive simulation software options, ensuring compliance with strict security standards like ITAR and SOC 2. It supports various deployment models and integrates with commercial schedulers, providing tools for administrators to manage security and user settings. The transition to cloud HPC involves addressing cost and security concerns, with Rescale facilitating the move to on-demand software licensing with application providers.

Rescale, a prominent player in computer-aided engineering simulation and high-performance computing (HPC), has been recognized by CIO Review as one of the "20 Most Promising HPC Solution Providers" in 2015. This accolade highlights Rescale's expertise in offering secure, scalable, and flexible HPC solutions, which have contributed to its significant growth and the establishment of the largest hardware footprint in the industry, supported by over 30 advanced data centers worldwide. Rescale provides a vast selection of over 120 simulation software solutions, including Siemens PLM and ANSYS, and prioritizes security and compliance with strict industry standards. The company’s robust platform allows enterprises, including Fortune 500 companies in sectors such as aerospace and automotive, to deploy simulation software securely through a browser interface, benefiting from the competitive advantage of scalable HPC resources.

In high-performance computing (HPC) environments, running solvers in a headless mode is crucial, which often involves executing batch processes via command-line instructions. At Rescale, solvers like OpenFOAM, Star-CCM+, and LS-Dyna can be run headlessly, but challenges arise when these solvers retain desktop GUI artifacts, necessitating the use of virtual frame buffers (Xvfb) to simulate display environments. Although Xvfb allows these solvers to function in non-headless conditions, it can lead to issues like race conditions and solver hangs, particularly during parameter sweeps or design of experiment runs. These pitfalls necessitate robust scripting solutions to manage virtual frame buffers efficiently. Additionally, Xvfb provides utility in rendering post-processing graphics without a physical display, enhancing the visualization of simulation data. Despite the complexities, Xvfb remains a valuable tool in adapting desktop solvers for HPC use, and continuous efforts are made to improve its robustness and effectiveness.

CONVERGE™ from Convergent Science is a prominent computational fluid design simulation program used in engine design, and its performance on different core types was evaluated using Rescale’s cloud simulation platform. The analysis compared the performance and cost of Nickel (HPC+) and Iron (HPC InfiniBand) core types, using a benchmark job that simulates a curving soccer ball. Although Nickel cores were found to be faster and more cost-effective for small clusters of up to 64 cores due to lower network bandwidth requirements, Iron cores offered superior performance for mid-sized clusters with more than 128 cores, attributed to their higher network bandwidth capacity. This performance disparity is linked to the communication overhead and slow interconnection experienced by Nickel cores at higher core numbers, highlighting Iron's efficiency for larger-scale simulations despite its higher cost.

CIOs face the challenge of creating a lean and agile IT structure that meets both current and future needs without incurring unsustainable costs, and cloud solutions offer a promising alternative to on-site HPC clusters. While traditional on-premise systems struggle with predicting and accommodating dynamic engineering demands, cloud platforms provide elasticity and a pay-as-you-go model that reduces capital expenditures and shortens procurement cycles. By transitioning to cloud-based HPC environments, companies gain better control over resource allocation, improve product development cycles, and enhance security through multi-layered protections, such as SOC2 compliance. Additionally, the adoption of cloud solutions minimizes the need for extensive IT staffing, allowing organizations to focus on recruiting top talent for critical roles. Despite some reluctance, many organizations, including Fortune 500 companies, are transitioning to cloud HPC systems to leverage economic, efficiency, and productivity benefits.

This article explores the challenges and debates surrounding REST API versioning, specifically focusing on managing API code to support different schema versions. It highlights the lack of consensus on the best method for specifying API versions, whether through URLs, headers, or other means, and the limited guidance on handling these versions within code. The author discusses an approach used by Stripe involving transformation pipelines that ensure core API logic operates with the current version, while compatibility layers handle version differences. The article further proposes a potential implementation using Django Rest Framework (DRF), leveraging Serializers to apply ordered transformations, ensuring backward compatibility. While this method isolates versioning concerns and simplifies core logic management, it acknowledges potential performance issues and challenges in handling complex breaking changes.