Activeeon, HPE and Intel have been working together on a seamless HPC and AI workload management solution for converging HPC and AI workloads (as announced in San Jose, California, December 9, 2020 HPE Press Release). Given that both workloads are very similar in their requirements, providing the right level of orchestration enables achieving the best efficiency and resource utilization.Download whitepaper
As high-performance computing (HPC), machine learning is not a finality on its own and must serve the interests of the company. Despite tooling and procedures being different from HPC to AI scientists, their prerequisites in terms of hardware are somehow identical: both HPC and AI workloads require access to large compute clusters using low-latency and fat interconnect as well as a large pool of storage.
In order to avoid system administrators and engineers creating inefficient silos on top of compute resources and storage, HPE, Intel & Activeeon have elaborated a solution architecture that manages and operates HPC and AI requests seamlessly. It is available with the “Seamless HPC and AI Workload Management” whitepaper by Activeeon, HPE and Intel. The document highlights key components features required to set up a complete and evolutive architecture for HPC and AI as a service needs.
In this HPC and AI workload management architecture, Activeeon scheduling & orchestration solution is a central component providing its powerful platform with user and developer portals, advanced pipelines and workflows with data connectors for distributed workload scheduling and meta-scheduling, as well as the management of compute resources. Activeeon scheduler and resource manager interact with native HPC clusters and native batch schedulers such as SLURM, LSF, PBS, acting as a meta-scheduler.
The whitepaper also highligts the way applications are packaged to offer insights about the importance of containerization into new deployment approaches, including complex hybrid environments. With the move to the cloud, application modernization is considered to be a priority to a growing number of businesses, the objective being to provide better scalability, resiliency and performance.
Containerizing HPC and AI applications requires deep knowledge to ensure bare-metal performance level is met. Those applications rely on specialized hardware environments, with accelerators, low-latency interconnect and parallel file systems. Accelerators like Field Programmable Gate Arrays (FPGA) are bringing a large set of advantages, as they can be reprogrammed and updated with new algorithms for different workloads. This flexibility allows a single FPGA to accelerate a wide range of different workloads efficiently, and to support future applications without the need to change the hardware.
As part of this common solution based on modern workload diversity, HPE has teamed up with Intel to offer FPGA solutions to answer the need for architectural diversity. And because no single architecture is best for every workload, XPUs, including CPUs, GPUs, FPGAs, and other accelerators, are also required to extract high performance. Here again, Intel® oneAPI products deliver the tools needed to deploy applications and solutions across these architectures.Download whitepaper
HPC customers who largely use on premises data centers are exploring cloud computing to improve the end-user experience, agility, and economics. In 2018, HPE launched the Early Access Program to support HPC cloud environments for customers and service providers, providing them access to demos and trials before buying. One of the objectives of the EAP is to explore End User (batch or interactive) or IT administrator experience and try to find the best solution for End User in HPC or AI environment.
Recently, HPE has decided to add a new dimension to the EAP by offering HPC as a Service (HPCaaS). It will be a unique offering that combines the capabilities of HPC GreenLake offerings, PointNext A&PS capabilities and provides the users with a compelling HPCaaS model.
The solution will incorporate building blocks from both VMaaS and MLOpsaaS and add specific HPC components that are required to complete the whole product. Users will be able to get access to HPC workloads seamlessly in a self-consumption model, and the batch users - to submit batch jobs that are portable, easier to build and deploy.
Achieving optimization of the HPC Experience has been made possible for HPE thanks to a close collaboration with a dedicated HPC Partner Ecosystem to help enterprises easily deploy solutions for any workload need, in any data center environment. As part of this ISV partners ecosystem, Activeeon is a key solution provider to optimize a range of software application needs: from automation to artificial intelligence. Activeeon solutions for AI and HPC are available with with the new HPE GreenLake cloud platform offering for HPC-as-a-service.
According to HPCwire, New HPE GreenLake cloud services for HPC with its set of pre-configured HPC services is planned for general availability in spring 2021.
Published on December 11, 2020