Velocity • Durability

VDURA CEO sees the end of the all-flash array era

VDURA CEO sees the end of the all-flash array era

Source: Block and Files : VDURA CEO Ken Claffey reckons that flash price rises and shortages have finished off the flash-will-kill-disks story and that tiered SSD and disk drive storage with tape for archives is the way to go – like the hyperscalers.

In a blog, entitled “The Flash Emperor Has No Clothes,” Claffey reacts to Everpure CEO Charles Giancarlo’s letter to customers, writing: “It is … the moment the Flash Emperor walked through town” with no clothes.

His starting point is that the SSD-will-replace-disk hype was not justified. At its extreme, we saw a Pure Storage exec predicting, back in May 2023, that no more hard disk drives would be sold after 2028 because of electricity costs and availability, as well as NAND $/TB declines.

Flash forward, as it were, and the then 4 to 5x price premium of SSD over HDD capacity has widened dramatically to 22.6x.

SSDs are not going away. Claffey writes: “Flash is the right medium for hot data, for performance tiers, for metadata, for checkpointing, and it always will be. This is a critique of an architectural pitch that bet your entire data estate against a commodity that was never going to stay cheap and that the people making the pitch never controlled.”

He thinks the pure all-in, store-everything-on-flash story was never true; “and Everpure’s letter which raises prices 70 percent in a year on top of an already-large gap has just removed the last fig leaf. The Flash Emperor has no clothes.”

Claffey says the hyperscalers never bought in to the all-flash idea, having internal 3-tiered storage spanning SSDs, HDDs and tape. He bangs te tiering drum: “Google’s Colossus is not all-flash. Meta’s storage backbone is not all-flash. Microsoft Azure’s storage is not all-flash. Amazon S3 and the bulk of EBS are not all-flash.”

“Every single one of them runs a mixed-fleet, software-defined architecture. Just enough NVMe flash to saturate the workload, then HDD for everything that doesn’t need flash speed,” and tape for the archival data.

In his view: “They understood that flash is a performance medium, not a capacity medium, and that pricing your infrastructure against a single commodity you don’t manufacture is a strategic error you only get to make once.”

He bangs his drum again: “Roughly 90 percent of the bill of materials in a modern all-flash array is the SSD or raw NAND itself. Everpure does not make NAND. VAST does not make NAND. WEKA does not make NAND. None of them controls its cost, its supply, its allocation, or its roadmap.”

They made flash cheaper through compression and deduplication, but it was never cheap enough to replace disk, and now “flash will not be cheap for years.”

The future is mixed-fleet platforms, with SSD and disk, “designed from the ground up to run heterogeneous media as one system. … This is the architectural pattern Google’s Colossus, Meta, and Microsoft’s storage backbones are built on.” And, Claffey says, VDURA. Follow their lead.