putilov_denis - stock.adobe.com

IBM releases compact, ultra-fast storage for AI

IBM released the elastic storage system 3500, a new object storage array that uses Spectrum Scale to provide fast storage for AI modeling and training.

IBM expanded its elastic storage system line with a new array that outperforms the rest of its software-defined portfolio through a combination of new hardware bundled with IBM Spectrum Scale.

IBM released its new elastic storage system (ESS) 3500 last week. It can provide up to 91 GBps in sustained throughput for overcoming storage barriers associated with AI and machine learning workloads, according to IBM. It also uses Spectrum Scale to increase performance through concurrent access and aggregate multiple types of storage.

The new distributed object and file array can also house up to 1.1 petabytes in 2U, which can help organizations consolidate high-end workloads.

While IBM positions the ESS 3500 as ideal for AI workloads, Scott Baker, CMO of IBM Storage, said the array could be used for workloads in media and entertainment, content management systems and streaming videos.

What's in the box

The ESS 3500 uses the AMD EPYC 7642 processor and NVMe storage. The flash storage relies on PCIe Gen 4 SSDs that, along with the processor, provide a throughput of up to 91 GBps, according to the vendor's spec sheet. IBM also stated the array can scale from 46 TB to over 1 PB using a 2.5 compression rate; when scaling the ESS 3500 to a cluster of 20 units, the sustained throughput can be increased to 1.8 TBps.

The ESS 3500 comes with impressive performance claims for a single 2U box, according to Brent Ellis, a senior analyst at Forrester Research.

"[IBM's] stated performance is about twice as fast as a similar Vast system," Ellis said. Other systems could provide faster performance but require a larger footprint, he added.

The 3500 targets data analytic workloads, which need storage that can quickly ingest large amounts of data without saturating the Fibre Channel connections, Ellis said. To avoid this, one would need either 100 Gigabit Ethernet or 200 Gigabit Infiniband, both of which are in the ESS 3500.

Another way to think about the ESS line is that it's like a powerful just a bunch of disks (JBOD) with software-defined storage bundled together, creating a storage system that can be used to manage other elements of a single data fabric, Ellis said.

IBM ESS 3500 front
The front view of the IBM ESS 3500.
IBM ESS 3500 back
The rear of the IBM ESS 3500 shows the network connectivity.

Leaning on Spectrum Scale

Spectrum Scale is also part of the ESS 3500, which makes it a fairly interconnectable product, Ellis said. The software-defined storage establishes a global data platform, which can aggregate together disparate data from across the organization and create a caching layer, Baker said. It can also sit on top of other storage systems, allowing users to manage existing infrastructure through the ESS system. The data is not moved, but a copy of the data is made available for access for high-end workloads.

Spectrum Scale also brings S3 compatibility to the ESS 3500, enabling connectivity to cloud-based storage environments, Baker said.

The bundling of a preconfigured Spectrum Scale on the ESS 3500 can be beneficial to customers that don't have a storage specialist to run it, Ellis said.

"Essentially, [IBM is] bundling two products, preconfiguring it for you so that you can start using it as soon as possible," he said.

Spectrum Scale is based on IBM's general parallel file system (GPFS), which is used in its mainframe units. The older software stack meets current needs but may need a redesign at some point in the future, Ellis said.

Adam Armstrong is a TechTarget news writer covering file and block storage hardware, and private clouds. He previously worked at StorageReview.com.

Dig Deeper on Flash memory and storage

Disaster Recovery
Data Backup
Data Center
and ESG