With businesses beginning to rely on analytics, machine and deep learning (ML and DL), and artificial intelligence (AI) to run daily processes and generate competitive advantage, high-performance computing (HPC) is becoming more applicable outside of its traditional niche use cases, such as supporting academic and scientific research.
HPC is becoming more mainstream, but many IT professionals need help getting started. Architecting and implementing an HPC solution is a complex task. HPC workloads require capturing and processing massive amounts of streaming data in real time. Doing so cost effectively requires carefully designing the infrastructure to optimize throughput and input/output operations per second (IOPS) while minimizing latency. Additionally, testing these platforms to guarantee they will work as required under the stress of running a demanding HPC workload is an expensive, time-consuming and challenging process.
Dell EMC aims to simplify this process by productizing pre-configured and pre-integrated “Ready Solutions” that are optimized for specific HPC use cases and verticals. Currently, it offers Ready Solutions for HPC for the life sciences, digital manufacturing and research industries, as well as for the Network File System (NFS) storage access protocol and Lustre open source, parallel file storage that are commonly utilized for HPC workloads.
Dell EMC’s Ready Solutions for HPC can run on Dell EMC’s PowerVault ME4 storage arrays, which are designed to be dense, carrying 75% more drives and increasing raw capacity by 122% compared to the previous generation MD3 arrays for up to a total of four petabytes (PB) of capacity, per Dell EMC. Dell EMC also claims that the ME4 can achieve up to 320K IOPS and bandwidth performance of 5500 megabytes (MB)/sec, in large part by supporting a variety of fast-performing back-end storage networking, including 16Gb Fibre Channel, 10Gb iSCSI and 12Gb SAS. For dynamic scalability to meet changing compute and storage capacity requirements, the ME4 is a modular architecture.
In addition to running multiple ME4 arrays in parallel, Dell EMC customers also have the option of deploying their HPC workloads on Dell EMC’s Isilon scale-out network-attached storage (NAS) platform. Scale-out NAS solutions like architecture are becoming more popular for HPC workload hosting because they can start small and scale capacity, compute and bandwidth independently. According to Dell EMC, Isilon can scale to 144 node clusters of nearly 80 PB of capacity with an inline data deduplication ratio of 3:1, and that deliver up to nine million IOPS and 540 gigabits per second (Gbps) of network bandwidth performance.
Dell EMC’s HPC solutions are configured per best practices that Dell EMC gleans through its HPC and AI Innovation Lab. Through the HPC and AI Innovation Lab, Dell EMC engineers obtain early access to partners’ new technologies. In collaboration with these partners and computer science subject matter experts, Dell EMC develops and benchmarks new solutions. Notably, the lab contains Dell EMC’s Zenith supercomputer, which has more than 27,008 cores and is consistently ranked as a top 500 supercomputer worldwide. In aggregate, the lab has more than 10 PB of storage for collaboration. Customers also have access to the lab, to evaluate how their applications will run on particular solutions, and to tap into Dell EMC and partner expertise for how to optimize their infrastructure according to their particular workload requirements.
Dell EMC combines heritage strengths around facilitating large enterprise-caliber data lakes and cost-effective and scalable compute that stand to serve HPC workloads well from a technological stand point. As they become less specialized and more general purpose, HPC workloads will need to tie into broader data center environments that must support a heterogeneous range of workloads with agility. Storage Switzerland anticipates that IT professionals going forward will find additional value in tapping Dell EMC’s composable infrastructure capabilities, which creates virtualized fabrics of compute, storage and networking capabilities that can be composed based on application needs, to address this need.
Finally, getting the technology right is only half the battle in successfully introducing HPC as a business driver. IT professionals need to be in sync with business needs that will be met via HPC, and they need to have strong proof of concept to justify resulting investment. Dell EMC’s focus on maintaining close touchpoints to customers, for example through the HPC and AI Innovation Lab, can help in this regard.