An update to IBM's FlashSystem line of all-flash arrays this week beefed up storage density with a new custom hardware module design and introduced AI-driven data placement within and across arrays.
Three new IBM FlashSystem arrays will ship March 6:
FlashSystem 5600, which supports up to 2.5 effective petabytes (PBe) per 1U chassis and up to 2.6 million IOPS.
FlashSystem 7600, which supports up to 7.2 PBe in 2U and up to 4.3 million IOPS.
FlashSystem 9600, which scales up to 11.8 PBe in a 2U system and up to 6.3 million IOPS.
The new arrays include FlashSystem.ai, management software that uses AI agents to perform ransomware detection and automated recovery, and to calculate compression and data deduplication using field-programmable gate array (FPGA) processors inside IBM's custom FlashCore Modules. FlashSystem.ai also performs proactive performance tuning, including optimized data placement at the individual drive level, and non-disruptive data migration across fleets of arrays, such as third-party storage systems virtualized behind a FlashSystem controller.
"Non-disruptive migration of data from one array to another, when coupled with the AI capabilities, improves the options that the AI can present to address concerns over performance or capacity issues," said Scott Sinclair, an analyst at Omdia, a division of Informa TechTarget. "The AI can simply offer to move the workload over to another system and automate the process."
FlashSystem.ai provides recommendations to human operators before changes are made, including a "view rationale" button that displays the reasoning behind its conclusions.
Agentic storage management table stakes
IBM is not the first to offer automated storage management embedded within arrays. NetApp has AI Data Engine, which performs in-place data curation, real time synchronization and policy-driven governance within its storage systems. Dell has CloudIQ for its PowerScale NAS systems, which offers performance anomaly detection and impact analysis.
Unlike those systems, however, which focus on file and object storage, IBM FlashSystem is a block storage system, typically used with structured data applications such as core banking applications and databases. Thus, its AI automation could be a better fit for that kind of enterprise workload, according to Brent Ellis, an analyst at Forrester Research.
We expect it to happen more broadly in overall IT infrastructures, predictive movement of workloads between entire environments, with the help of AI.
Natalya YezhkovaAnalyst, IDC
"IBM is trying to address operations and scalability issues -- not necessarily the same problem that NetApp is trying to address, which is essentially easing the deployment of AI workloads," Ellis said. "IBM fits into a trend of increasing data security and auditability services inside storage platforms that enable the use of agentic AI operations and more flexible, automated control."
In general, however, real time AI-driven storage management is becoming table stakes for storage vendors, said Natalya Yezhkova, an analyst at IDC, replacing the previous rules-driven approach to performance optimization.
"Moving workloads among storage tiers has existed for many years, but dynamic placement according to SLAs [service-level agreements] and the health of the system is important," Yezhkova said. "We expect it to happen more broadly in overall IT infrastructures, predictive movement of workloads between entire environments, with the help of AI."
Brainy FlashCore Module sets IBM apart
IBM's custom-designed FlashCore module now supports up to 105 TB capacity.
In terms of raw capacity per storage enclosure, the new IBM FlashSystem arrays leapfrog Dell's PowerStore Q update last month, which packed 2 PBe into 2U using 30 TB QLC SSDs. However, these updates represent two different approaches to hardware design -- Dell was flexing its supply chain muscles by sourcing a high-capacity third-party SSD, while IBM builds its own custom FlashCore Modules, Ellis said.
"FlashCore Modules are physically larger than commodity SSDs, but this is the benefit of creating your own module, because IBM is the one that is organizing the flash NAND on the actual wafer, and so it can do some engineering and architecture to use the space better," Ellis said.
"It means IBM can source flash from multiple vendors and is cutting out a layer of the supply chain, which cuts out a layer of cost, and it allows it to more confidently be able to provide supply to customers," he said.
In Ellis's view, IBM FlashSystems' closest competitor is Pure Storage, which offers Pure1 AIOps, including automated workload placement across storage systems, and builds its own DirectFlash Modules that support up to 150 TB per drive.
But there are some differences between IBM and Pure's approaches that might be significant for some security-sensitive workloads, Ellis said. While Pure focuses on system-level storage management and resilience, IBM distributes storage and security management across the FPGAs in FlashCore Modules.
"IBM can run some of those functions directly on the drive, so essentially, you have security down to the drive, not just to the OS, and so you have a deeper root of trust boundary," he said.
Beth Pariseau, a senior news writer for Informa TechTarget, is an award-winning veteran of IT journalism. Have a tip? Email her.
