IBM Pushes Storage Limits With Storage Scale 6000
December 4, 2023 Alex Woodie
IBM recently unveiled a new high-end all-flash storage array that pushes the limits of I/O. Dubbed the IBM Storage Scale System 6000, the new array gives users 1.44 PB of storage in a 4U chassis with up to an eye-watering 256 GB/sec of bandwidth and 7 million input/output operations per second (IOPS) of performance.
Few IBM i shops have the need for that sort of performance with traditional business applications. Indeed, with its Storage Scale parallel file system (the new name given to IBM’s General Parallel File System, or GPFS, which also previously went by the name Spectrum Scale), the Scale System 6000 is being positioned primarily to power HPC, AI, and advanced analytics workloads.
But there’s no telling what individual IBM i shops are planning for their future data and IT systems. IBM is pushing the Scale System 6000 as the cloud backbone for a “global data platform” and as part of its “edge to core to cloud” strategy. With the capability to scale up and to scale out, IBM sees the Scale System 6000 moving lots of data out on the edge, or as a centralized platform where companies consolidate data from multiple disparate storage systems (file and object-based), where it can be poked and prodded to reveal greater value.
The new Storage Scale 6000 all-flash array boasts data throughput of 7M IOPS.
For instance, IBM says the Scale System 6000 could house data used for medical research, to analyze data related to customer feedback and buying patterns, to detect potential security threats, or to improve manufacturing processes. In that respect, it’s absolutely conceivable that the new array could play some role housing data originating in midrange Power Systems boxes. It all depends on the data and AI strategies of IBM customers, which are evolving at lightspeed at the moment, as they are for most of the world’s large businesses. Scale System 6000 simply gives them more options at the high end of the spectrum.
The Scale System 6000 also showcases what IBM can do with storage tech, which is also pertinent to readers of this newsletter. There’s no denying that the specs and capabilities of the Scale System 6000 are impressive.
Powering the array are dual 48-core AMD “Genoa” Epyc storage controllers in an active-active configuration, connected to 768 GB to 1.5 TB of RAM. Each 4U node can be outfitted with either 24 or 48 NVM-Express drives with about 4 TB to 30 TB in capacity. Alternatively, users will soon have the option of using 24 or 48 NVM-Express FlashCore Modules storing 19 TB to 114 TBe (or terabyte equivalent, with inline compression).
For the system connection, IBM offers up to 16 ports of 100 Gb/sec Ethernet with support for low latency RDMA over Converged Ethernet (RoCE), 200 Gb/sec HDR InfiniBand, or 200 Gb/sec or 400 Gb/sec NDR InfiniBand, or a combination of any of these. The ultimate goal with all this fast network connectivity is to keep Nvidia GPUs fully saturated with data to optimize the training of machine learning models.
The Scale System 6000, which runs the Red Hat Enterprise Linux, can be scaled vertically or horizontally. Thanks to the parallel file system, users can link thousands of Scale System 6000 nodes, delivering a whopping 633 yottabytes (YB) of total storage capacity in a single cluster. (Of course, there are likely very few applications in the world that require that much data to be housed in a single namespace, but IBM wants you to know that it’s there, if you need it.)
IBM envisions the Scale System 6000 as a key component of the Global Data Platform.
The Scale System 6000 stores data in IBM’s proprietary parallel file system, dubbed Storage Scale, but it has the capability to expose that data in any number of different file protocols, including NFSv4, SMB, HTTP, S3, HDFS, POSIX, Nvidia GPUDirect, Container Native Storage Access (CNSA), and Container Storage Interface (CSI). IBM supports two- and three-site data replication schemes (asynchronous and synchronous), as well as IBM’s proprietary RAID-like erasure encoding tech.
The real selling point of the Scale System 6000, however, is speed. With the capability to move 256 GBps over PCIe5 links in its “turbo tier” mode, the Scale System 6000 delivers twice the throughput of its smaller brother, the Elastic Storage System 3500, which is based on PCIe generation 4 tech. The ESS 3500, which tops out at 24 NVMe drives, boasts eight InfiniBand or Ethernet adapters compared to the Scale System’s 16 PCIe adapters to drive up to 48 NVM-Express drives per node. The Scale System 3500 also has slower 200 Gb InfiniBand and 200 GbitE, compared to the 400 Gb/sec InfiniBand and Ethernet (RoCE) links.
All that high-end networking and storage tech means companies can move a heck of a lot of data very quickly. And if a customer is feeling limited by 7 million IOPS, they can crank it up to 13 IOPS with the “turbo” mode, IBM says.
With SS 6000 deliveries set to start December 6, IBM is claiming bragging rights in the high-end all-Flash segment, which also features systems from Dell, DDN, Huawei, NetApp, Pure Storage, and WekaIO. “We did it again,” IBM’s David Wholford wrote in a blog. “We are the 2.5 times the throughput performance than marketing leading competitors.”
As the launch of ChatGPT one year ago continues to drive the generative AI revolution to new heights, IBM is keen to capture a share of the market with hardware like the Scale System 6000 and software, via its watsonx line. With an array like the Scale System 6000 pumping out data to high-end GPU systems, like Nvidia’s DGX line, IBM hopes to give customers the tools to train their own GenAI models.
“The potential of today’s new era of AI can only be fully realized, in my opinion, if organizations have a strategy to unify data from multiple sources in near real-time without creating numerous copies of data and going through constant iterations of data ingest,” IBM Storage general manager Denis Kennelly stated in a press release. “IBM Storage Scale System 6000 gives clients the ability to do just that – brings together data from core, edge, and cloud into a single platform with optimized performance for GPU workloads.”
You can read the Scale Storage 6000 announcement letter here. You can read the data sheet here.
RELATED STORIES
Direct Attached Storage Gets Massive NVM-Express Expansion
Storage Helps Power Systems Grow In The First Quarter
IBM Unveils New FlashSystem Storage Arrays
