Facilitate HPC Deployments with Reference Designs for Intel Scalable System Framework

Print Friendly, PDF & Email

Sponsored Post

intel_CircleIntel has talked at length about Intel Scalable System Framework (Intel® SSF) as its design approach to scalable, balanced HPC systems built on processors, fabrics, storage, and software specifically designed to work in concert. Now, with Intel Scalable System Framework Architecture Specification and Reference Designs, the company is making it easier to accelerate the time to discovery through high-performance computing. The Reference Architectures (RAs) and Reference Designs take Intel Scalable System Framework  to the next step—deploying it in ways that will allow users to confidently run their workloads and allow system builders to innovate and differentiate designs based on Intel Scalable System Framework .

The RA describes the high-level system architecture and application platform. A Reference Design is one verified implementation of a Reference Architecture. The design includes hardware and software Bill of Materials information and cluster configuration instructions that can be used “as is” or built on.

Andy Parma, Intel Marketing Manager, explains: “The Reference Architecture and Reference Designs aren’t simply concepts. They are documented, validated deployment examples based on industry best practices and conventions to give specific types of workloads, such as modeling and simulation, the high-performance resources they need for advanced computing. Intel has built and validated systems based on the reference designs.”

Over 30 companies have committed to Intel SSF, including leading system builders and ISVs, who will validate their products. Dozens more are expected to engage with Intel with the transition of the Intel® Cluster Ready program partners to the Intel Scalable System Framework program.

While the Reference Architectures define a system that at a minimum has a high-performance Intel processor and interconnect and validated HPC tools and software,” Parma adds, “it leaves room for the system builders to design to their customers’ requirements and to differentiate and innovate on their own. For example, Intel® Enterprise Edition for Lustre* Software is part of Intel SSF, but some customers will have other file system requirements and thus will choose a different technology. Also, it gives builders the options to choose Intel® SSDs or spinning hard drives, again, depending on customer needs and design requirements. So, there’s built-in flexibility, but also foundational requirements to make it a high-performance Intel SSF system.”

What these documents mean to the system builders is that they can work from a proven foundation of technologies to deliver high-performance solutions quickly and enhance and modify the design to meet customer needs. To the end user they mean faster time to insight with the knowledge that software from Intel partners will run on Intel SSF systems.

The first Intel SSF configurations target classic HPC clusters, used in both enterprise and research/academia, which support traditional modeling and simulation workloads. Two Reference Designs cover cluster builds, one based on Intel® Xeon® processor E5-2600 v4 family with Intel Omni-Path Architecture fabric and the other based on Intel Xeon processor E5-2600 v4 family with Intel® Ethernet controllers. Both are based on the OpenHPC* v1.1 software stack.

These documents are only the beginning of what will grow to additional RAs and validated Reference Designs based on the architectures. In the future, additional Reference Architectures will span small clusters through scalable supercomputers, systems that require special capabilities not found in smaller configurations. RAs will be written that support cloud, data analytics, machine learning, and visualization workloads in addition to traditional HPC workloads.

What’s important here is the compatibility of software across the architecture as solutions evolve,” commented Parma. “Users want to have a consistent framework that can run their many workloads without having to deploy an entirely different set of components. Intel SSF and the Reference Architectures give them that.”

Introduced at the same time as the reference architectures, Intel® HPC Orchestrator family of products include a configuration of pre-integrated, pre-tested, and pre-validated system-level software and tools with ongoing support by Intel and ready for deployment on a Reference Design or other HPC cluster. It is an HPC software stack central to the Intel SSF approach and built on the OpenHPC community system software, hosted by the Linux* Foundation. In addition to the OpenHPC software, it adds some proprietary Intel HPC tools, such as the Intel® Parallel Studio XE Cluster Edition 2016 Suite. Like the Intel SSF configurations, Intel HPC Orchestrator is designed to reduce the time required to deploy an HPC solution and enable greater democratization of HPC for a wider community of users.