No small storage company has had more press coverage and “buzz” than “ioMemory” maker, Fusion-io. I have long marveled at the company’s ability to attract attention, but this has rubbed some analysts wrong. As component vendors enter their space, how can a premium company with proprietary products compete over the long term?
See all those NAND chips? The companies that make them are competing with Fusion-io. So how can they survive?
The Facts of the Flash Market
Fusion-io sells “accelerator” and “memory” cards packed with NAND flash memory. Flash is perhaps the hot commodity in enterprise (and consumer) storage right now, and is changing the shape and architecture of everything from portable devices to the datacenter.
Fusion-io has become the standard-bearer for the use of flash as ultra high-performance memory rather than in the form of a solid state drive (SSD) masquerading as a hard disk. Only one other company (Virident) is focused on PCIe flash memory cards like Fusion-io, though many are looking to enter the market thanks to the NVMe and SCSI Express initiatives. Pretty much every other product on the market today (including other PCIe cards) relies instead on the same block storage interface as an SSD.
Fusion-io touts the performance and reliability advantages of doing away with the SCSI storage interface and presenting flash as memory. After all, with less abstraction and protocol conversion in the way, “ioMemory” has got to be better, right? But real-world tests show that PCIe SSDs from companies like Micron, Violin, Texas Memory Systems, and others offer excellent performance, too.
One must also consider the fact that Fusion-io doesn’t manufacture the NAND flash chips used by their cards. These costly chips certainly make up a large part of the cost of any flash-based storage product, and companies like Fusion-io are at the mercy of their suppliers, mainly Micron/Intel and Samsung. Yet Micron, Intel, and Samsung are aggressively moving into the PCIe and SSD market with their own consumer-ready products!
How can a premium product like Fusion-io’s famous ioDrive compete with pretty-much-as-good PCIe SSDs from the very companies that produce their NAND flash components? Some have become downright bearish on Fusion-io, predicting imminent failure for the company. I’m not one of those people, however.
A Changing Market
Fusion-io has a few solid cards yet to play. Their core value isn’t the cards themselves, it’s the flash controller and interface techniques that make these chips accessible. This is an incredibly important element of any NAND flash product, and the reason Apple bought Anobit, LSI bought Sandforce, and Marvell and Samsung are looking so good right now.
In short, NAND flash is useless without a good controller, and Fusion-io has one of the best in the business. The entire industry appears to agree that memory-oriented access will play a growing role in solid state storage, and Fusion-io is at the lead of this technology. Their controllers have proven reliable, extending the life of NAND chips and standing up to enterprise workloads.
This is why Fusion-io is the supplier of choice for enterprise flash efforts from IBM, Dell, HP, Cisco, and NetApp as well as high-profile upstarts like Nutanix, Kaminario, NexGen, and others. These companies could easily find cheaper suppliers, but they have chosen Fusion-io.
Although Fusion-io must be a little reluctant to share this market, they’ve even jumped into the SCSI Express effort with the SCSI Trade Association, though they are keeping their distance from the competing NVM Express initiative.
It is important to note that, even if SCSI Express and NVMe are wildly successful, these standards don’t spell doom for providers of novel controller technology. On the contrary, these interfaces require excellent implementation to succeed. And they don’t obviate the creation of differentiating enhancements “on the back end”.
PCIe memory products don’t look or act like a hard disk drive, though a software layer can present them to the operating system that way
Fusion-IO ioScale and Open Computing
But Fusion-io’s success goes beyond their controller. One of the biggest markets for Fusion-io’s products are massive Internet services offered by companies like Facebook, Apple, Google, Amazon, and so on. Users demand instant response to web queries, and solid state devices like Fusion-io cards are the only way to meet that need.
There was much speculation about which product companies like Facebook were using for their storage, but that was cleared up yesterday when it was revealed that Fusion-io is a major supplier there. This is one area where the product’s performance is probably a secondary concern: The direct interface to Fusion-io’s cards also cuts out numerous points of failure (RAID controllers, drive interfaces, power supplies) that could compromise reliability.
Fusion-io also revealed that they are producing a special high-volume, low-price card for this market. The new ioScale card appears to share technology with the low-cost ioFX card introduced last year. But the ioScale is truly massive, with directly-accessed capacity up to 3.2 TB. This compares favorably to other solutions that rely on RAID to combine multiple devices for that kind of scale.
3.2 TB of flash and a single controller chip. That ioScale is one hot PCIe drive!
Facebook Giveth and Taketh Away
Alongside the announcement that they use Fusion-io cards, Facebook announced at the Open Compute Project Summit that they would ditch the majority of their OEMs and have their own hyperscale servers built to the Open Compute spec. This is a serious setback for Dell and HP (which had much success selling into hyperscale markets) since they are effectively cut out of this promising market entirely.
Open Compute also poses an issue for Fusion-io, since it’s clear that the big hyperscale players are tired of paying a premium for components. But the company has a clever response to this challenge, too: Fusion-io is “open sourcing” the design of the ioScale card.
Open Compute makers are free to build their own ioScale cards or integrate them into the server or storage devices. But Fusion-io didn’t open source everything: Companies must still buy Fusion-io controllers and use their APIs and interfaces!
This was really the only positive way to respond to Open Compute. After all, if they insisted on being an OEM, Fusion-io might have found themselves cut out of the market entirely. Instead, they stand a strong chance of becoming the core “tier 0″ storage component maker not just for Facebook but anyone else who adopts Open Compute!
Stephen’s Stance
Fusion-io has crafted an effective response to the reality of commoditization: They are focusing on their core value (the controller and interfaces) and working to become a standard part of the growing hyperscale server market by joining Open Compute. Although techies were likely dazzled by the capacity and price of the ioScale card, this strategic move is much more important to the future of Fusion-io!
Disclaimer: Apart from a few pins, a Woz-signed book, and a trade show party invite, I’ve never gotten anything from Fusion-io. They do regularly brief me about their products, however, as do most of the companies mentioned in this article. Except Apple; they never brief anyone!
