By Chris Mellor, Contributor
Enterprise flash drives will move to dominate tier 1 storage over the next 12 months to 36 months. Data centre storage arrays will see their 15,000 rpm Fibre Channel hard drives replaced by NAND enterprise flash drives that offer much faster access at a price that will be comparable to Fibre Channel disk. Does this strike you as likely? No? Let's try and argue the case for it.
Enterprise flash drive capacities set to grow
Violin Memory offers a 5 TB externally attached 1010 Memory Appliance today, one that uses single-level cell (SLC) NAND chips. It's more expensive than a Fibre Channel drive array and isn't going to occasion a mass transfer of tier 1 data to NAND enterprise flash drives. But, Violin is set to introduce 10 TB and 20 TB versions later this year and hopes to achieve a 40 TB version. With those capacities it starts to look interesting.
Violin is going to use Toshiba multi-level cell (MLC flash) and rely on its switch controller and multi-lane access to the NAND modules to provide fast write speeds and beat the limited write-endurance problem. DRAM caching is already used to cache writes so that host server applications won't have to wait for a write I/O to NAND to complete. Toshiba has invested in Violin Memory and agreed to a strategic supply deal with it. Conclusion: Toshiba is serious about this and the potential here is quite startling.
When we consider that Toshiba partners with SanDisk (which has 4-bit MLC patents), then we could see a 40 TB 2-bit MLC array jumping to 80 TB capacity with 4-bit MLC. Start aggregating these arrays together and 100 TB-plus enterprise flash drive storage becomes feasible.
Violin competitor Texas Memory Systems, which already makes externally connected flash arrays with capacity in the tens of terabytes, will no doubt follow suit in the capacity game.
To make the case stronger still, look at Fusion-io, a supplier of PCIe bus-connected flash cards with hundreds of gigabytes of capacity, using either SLC or 2-bit MLC flash. It knows what's coming on the Violin and Texas Memory Systems fronts, and it knows that an enterprise flash drive array removes the need for a flash card tier of storage in a server. It will either have to make flash storage arrays itself or bulk up its PCIe-connected cards to capacities of multiple terabytes and then tens of terabytes.
Fusion-io has already developed a flash-based SAN array, one that never came to market. This was ioSAN: a vanilla x86 server pre-loaded with the company's ioDrive cards, offering a sharable NAND flash storage resource. Such a networked flash storage array provides Fusion-io with a possible route to compete with Violin Memory and Texas Memory Systems. Note: Samsung, with is flash chip foundries, has taken an investment in Fusion-io.
Here's another nail to hammer into the Fibre Channel drive coffin. It's estimated that within the next 18 months to 24 months, Xiotech will introduce a flash-only Intelligent Storage Element (ISE). The hard drive-based ISEs currently offer terrific IOPS and bandwidth performance and have a five-year warranty with their fail-in place technology. Imagine what a flash-based ISE could do in terms of IOPS, bandwidth and warranty.
Currently, leading-edge flash is transitioning to a 25 nanometer process geometry but sub-25 nanometer processes are in development, and these will allow the likely doubling of flash capacity in storage devices. Thus, a 40 TB SLC array would become an 80 TB one, and an 80 TB 2-bit SLC array would increase to a 160 TB product, probably with a lower per-terabyte cost.
Enterprise flash drives: Impact on Fibre Channel suppliers
The argument being made here leads to the view that makers of 15,000 rpm Fibre Channel hard drives are looking market death in the face, and it's pretty much not a matter of "if" but of "when" enterprise flash drives will supplant them.
The "when" could be sooner than they think. Nobody wants to buy expensive, fast, Fibre Channel drive arrays; they want fast access to data, and if they can get that from reliable and blisteringly fast enterprise flash drives at a comparable price with lower electricity needs, then they'll move. They'll move in droves and get rid of a storage I/O bottleneck that's hobbling their 4-core, 6-core and coming 8-core, multi-socket servers that could run more virtual machines (VMs) than they currently do if only the storage could keep up with processor and networking developments.
What happens to enterprise storage array suppliers if tier 1 data migrates to flash? The ones that offer comparatively small flash caches in their arrays for hot tier 1 data will find themselves becoming overpriced and restricted in capacity. They will have to junk their expensive enterprise flash drive caches as well as their Fibre Channel hard drive enclosures and follow suit. This affects EMC, Hitachi Data Systems and IBM, which will each have to bring out all-flash enclosures for tier 1 storage.
For NetApp the effect will also be substantial. It currently offers an add-in flash Performance Acceleration Module (PAM) card for its controllers. This will be insufficient to compete with competitors offering solely flash tier 1 storage, and NetApp will have to introduce flash storage enclosures in its FAS unified storage products.
The effect on Fibre Channel hard drive suppliers could be dramatic, with unit sales collapsing and possibly drastic effects on profitability. Seagate would be much affected by this and possibly, even probably, looking to leverage its partnership with LSI to develop Engenio flash storage products using its own Pulsar flash technology.
Seagate is a rich and hungry animal, and it's not going to let itself starve because Fibre Channel drives are going out of fashion. It's certainly not going to want to let Western Digital make flash storage array hay without joining the party. The odds are that Seagate will transition its high-end Fibre Channel drive business to flash and the low end to 2.5-inch SAS drives.
So, flash is emphatically not a flash in the pan. The pricing, write performance and write endurance issues are being addressed and the solid-state drive industry knows there's a door ready to open in front of it here. These vendors will be knocking on it soon enough and those knocks will spell doom for Fibre Channel hard drives. The era of tier 1 data being stored on spinning rust is coming to a close. Case proven? We'll see.
Chris Mellor is storage editor with The Register.
This was first published in April 2010