Looking at the night sky, I always wonder how many galaxies, stars, and planets are out there. This predicament has baffled not only me but has challenged humankind for centuries. From the last many decades, scientists, and engineers in the sphere of cosmology and other sciences like fluid dynamics, drug discovery, genomics, and weather forecasting have been able to find answers to some of the most daunting questions using the computational power of supercomputers.
With continued innovation in computer hardware, access to more data, and sophisticated simulation software, scientists, and engineers have been able to create complex scenarios to theorize possible outcomes. To find answers faster, researchers, scientific communities, and businesses alike have been investing in building high-performance computing (HPC) capabilities in-house. That means managing their HPC data centers. Without doubt, these investments have paid dividends. According to Hyperion Research, an HPC investment results in an average revenue of $463 per dollar invested in HPC and an average profit of $44 per dollar for every dollar of HPC investment.
Most HPC workloads are run on-premises to achieve high performance, low latency, and high throughput. Organizations want to maximize return on investment, and so optimizing HPC workloads is critical because these machines come with an expensive price tag. Running HPC workloads on-premises comes with the following challenges:
Capacity planning: One of the biggest challenges that HPC customers face is capacity planning. Organizations either overprovision or underprovision capacity, leading to long queuing times impacting productivity.
Legacy hardware: Refreshing infrastructure generally happens in 3–5 years, meaning HPC system users have to wait to get access to the latest HPC infrastructure. That slows innovation and productivity.
Cost to manage data centers: Includes covering for power, cooling, networking, storage, hardware, and software. For most organizations running HPC workloads, managing facilities and hardware is not their core competency.
Security and compliance requirements: Legacy systems need regular maintenance to ensure that they meet the security and compliance requirements.
Increasingly, HPC workloads in the cloud have become a reality. According to a 2020 Intersect360 Research report, the cloud has seen double-digit growth in the last 5–6 years, with a 22.0% compound annual growth rate (CAGR) and a projected cloud segment to exceed $3.8B by 2024.
Figure 1: Consumption of HPC in public cloud ($B), Intersect360 Report.
Moving your HPC workloads to the cloud including the following benefits:
Better performance: Get access to the latest generation of infrastructure and on-demand capacity to address your business requirements. HPC customers can deploy bare metal virtual machines (VMs), GPU, and Compute shapes as needed.
Optimize costs: With the cloud, you don’t have to worry about upfront capital costs. Use the pay-as-you-use pricing model while running without interruptions.
Flexibility to scale: Cloud offers organizations the flexibility to scale up and down their IT requirements to fit the business's needs.
Focus on your core business: With HPC in the cloud, you don’t have to worry about building and managing your data centers, letting you focus on your strategic business priorities.
Oracle offers excellent value for HPC workloads in the cloud, helping organizations big and small run their critical HPC use cases without having to worry about substantial upfront investments.
Latest generation of computer hardware: Oracle has partnered with AMD and Intel to offer its customers the latest generation of compute capacity, which includes a wide range of VM and bare metal shapes. To help accelerate visualization and run complex data models for HPC and AI workloads, Oracle has partnered with NVIDIA and is offering Tesla P100, V100, and A100 GPUs. Hear more from our EVP Clay Magouyrk about Oracle's strategic partnerships with Intel, NVIDIA, and AMD.
High bandwidth, low latency network: Oracle offers low-latency, high-throughput networks to successfully run HPC uses cases in the cloud. Oracle Cloud Infrastructure (OCI) network is a non-oversubscribed, highly scalable network with approximately 1 million network ports in each availability domain, with high-speed interconnections and latency < 100μs between the hosts in an availability domain. Also, traffic between the availability domains is encrypted. For HPC workloads, Oracle offers high bandwidth, ultra-low latency RDMA network with latency as low as 1.5 microseconds.
Easy to deploy file systems: Oracle offers HPC customers an array of parallel file systems they can choose from: IBM Spectrum scale, BeeGFS, Lustre, and more. Customers can deploy HPC file systems on OCI in a few clicks.
Guaranteed service-level agreements (SLAs): Oracle is the only cloud vendor that offers end-to-end SLAs across availability, manageability, and performance. To learn more, see the documentation.
Industry-leading price-performance that either matches or exceeds the on-premises deployments. Check out the blog by my colleague Kristen Yang on Optimizing HPC workloads with Oracle Cloud Infrastructure on Intel.
Consistent global pricing: You can better manage your cloud technology expenditure with Oracle’s consistent global pricing.
Fast deployment of HPC workloads: Oracle makes it easy for you to deploy and scale HPC workloads in the cloud. Customers can automate the deployment of HPC clusters through Oracle Cloud Marketplace. To see how, watch the video.
Secure by design: OCI provides zero-trust, security-first architecture with easy-to-implement security controls.
Ecosystem of HPC partners: A wide ecosystem of HPC technology partners that help deliver improved and tailored solutions that meet HPC customer requirements.
Many of our customers like Nissan, YellowDog, DeepZen, GridMarkets, and Altair have successfully transitioned to the cloud and are reaping the benefits while getting superior price-performance. To check out more such stories and learn about Oracle’s HPC solutions, visit us at High-performance computing solutions on Oracle Cloud Infrastructure.