In today’s hybrid cloud reality, distributed data holds the key to unlocking value through new business and operational insights. However, according to the "Global AI Adoption Index 2021,"...
In today’s hybrid cloud reality, distributed data holds the key to unlocking value through new business and operational insights. However, according to the "Global AI Adoption Index 2021," conducted by Morning Consult and commissioned by IBM, data complexity and data silos are top barriers to AI adoption. Overcoming this is critical for improving data access and transforming business processes, from supply chains and asset management to analytics.
Addressing these challenges, we announced key enhancements to our next-generation software-defined storage, IBM ESS 3500. IBM ESS 3500 is designed to accelerate data delivery for AI workloads and help speed time to market with cloud-scale performance and capacity.
The growing adoption of AI and Kubernetes by enterprises requires a new model that simplifies data access, increases productivity and scales readily as these projects grow. IBM understands distributed file and object workload needs and the need to solve for myriad use cases, including design simulations, especially AI and ML.
Accelerating data science and optimizing application development
IBM ESS 3500 is engineered to help clients to accelerate data science, modernize and optimize application development, simplify and accelerate DevOps and optimize content repositories. IBM ESS 3500, enabled by Spectrum Scale, is designed to provide enterprise class security and availability with a global namespace supporting the unification of data from multiple sources across core, edge, and cloud without the need to make additional copies of data.
Michael Sedlmayer, Re-Store President, Supercomputer Architect, had this to say: “Our customers have been using the capabilities of IBM Spectrum Scale and the IBM ESS family. They are drawn to the value of data resiliency and security with IBM Spectrum Scale and ESS from accidental and malicious attacks that can result in loss of data. Our customers appreciate the power, and the way IBM handles hard to solve problems better than anyone else. We are currently transitioning our sales efforts to the new IBM ESS 3500 to reach new customers, as we have found nothing compares to the functionality and performance of the IBM ESS with IBM Spectrum Scale.”
Improve AI training time using IBM Spectrum Scale and IBM Elastic Storage Systems
IBM ESS 3500 is optimized for AI-accelerated computing solutions, such as NVIDIA DGX systems with GPUDirect support. Based on a recent client result, IBM can improve AI training time as much as 70% using IBM Spectrum Scale and IBM Elastic Storage Systems. The solution is designed for compute-intensive workloads with the ability to scale from 46TB to ~1PBe effective capacity in a 2U form factor using LZ4 compression with a 2.5x compression rate and is projected to support over 1.8TB/s in a 20-node rack configuration.
“AI workloads demand powerful infrastructure that delivers cost-effective performance and scale, which is why customers tackling the most challenging AI opportunities depend on NVIDIA DGX systems and NVIDIA DGX SuperPOD,” said Charlie Boyle, Vice President of DGX systems at NVIDIA. “Building on NVIDIA’s long collaboration with IBM, the new IBM ESS 3500 storage system enables DGX customers to quickly and easily scale their infrastructure to speed AI-powered insights from their data.”
IBM ESS 3500 will be generally available on May 20, 2022. For product details and a demonstration, IBM is hosting an on demand webinar (registration required) available for 90 days through August 15, 2022.
 Based on data from an actual customer who used IBM Spectrum Scale software and IBM ESS as compared to previous configurations using similar workloads. Disclaimer: Results based on one customer https://www.ibm.com/case-studies/continental-automotive/ All client examples cited or described are presented as illustrations of the manner in which some clients have used IBM products and the results they may have achieved. Actual environmental costs and performance characteristics will vary depending on individual client configurations and conditions. Generally expected results cannot be provided as each client's results will depend entirely on the client's systems and services ordered.
 Disclaimer: Effective capacity based on internal testing of compression technology of various workload types and is based on publicly available documents (https://github.com/lz4/lz4). Actual customer capacity may vary up or down depending on type of data that is stored.
 Disclaimer: IBM Spectrum Scale can “allow concurrent read and writes from multiple nodes.” Reference: https://www.ibm.com/docs/en/spectrum-scale/4.2.0?topic=gpfs-improved-system-performance and performance results for 20 nodes assume workloads can be spread evenly across all nodes in the configuration.