The Rise Of White-Box Storage

When EMC announced its ViPR storage gateway last year, its list of supported hardware included "white-box storage." In many ways, this is a tacit recognition of the changes the industry faces in storage hardware platforms. The implication of white box is low-cost, generic gear.

To understand the white-box storage trend better, let's examine traditional storage arrays. They all use similar components, but there is a proprietary twist to how these are used. For example, not only are the hard disk drives "enterprise-class" with SAS interfaces and other features, but they are sold only to OEMS and sometimes have proprietary hooks in them, such as identification codes. The result is a whopping markup on the drives, both by drive vendors and by the array providers.

Then along came SSDs. All SSDs are much faster than the fastest hard drives. This fact has uprooted the decades-old storage tiering of primary fast HDD and secondary slower HDD and replaced it with primary SSD and secondary bulk SATA HDD. The latter are currently around three cents a gigabyte, while SSDs are about 50 cents per gigabyte on the Internet, which is less than the OEM cost of a traditional enterprise HDD.

A note to avoid confusion: Most SSD to HDD comparisons are between SSD and cheap 1TB hard drives, which is typically the only cost data available. That ignores the huge performance difference and is the equivalent of comparing a Ferrari and a Ford Explorer. The correct comparison is between SSD and the fastest enterprise (SAS) HDD.

Pricing, and price/performance, wouldn't be enough to upset the apple cart, though. The other key technical factors at work are the very structure of the array and the migration to alternatives to RAID.

The base computing for an array (sometimes called the head node) has migrated to an x86 CPU base, though ARM has made a recent showing. This means the RAID head is essentially a server, and that lowers the technical barrier to entry almost completely.

At the same time, RAID is running into a wall of problems due to long rebuild times for very large drives. We are up to two parity drives (RAID 6) and now looking at erasure code systems capable of tolerating as many as six simultaneous drive failures.

As an alternative to RAID, new replication systems have garnered a huge market segment, especially in cloud computing. These object stores use cheap servers and drives in huge quantities and have excellent data integrity results.

The success of the cloud has highlighted how storage hardware is made. Large cloud providers such as Google, Amazon, and Microsoft have cut out the middlemen and now buy their gear direct from the Chinese ODMs that for years made boxes for the big-iron storage vendors. These CSPs are essentially replacing EMC and others with direct buying.

It's clear that the direct-buy method -- which is keenly watched by enterprises -- is stable and rugged enough for broader access. The ODMs are beginning to open up their own branded outlets in the US and selling through distribution. These white-box products are rarely actually white and unbranded; they constitute products from a set of suppliers competing mainly on price while generally offering the same features and even the same software. (Ceph software, for example, runs on any COTS hardware, including white boxes.)

Larger enterprises, especially those with a foothold in China, have looked to more direct ODM relationships, while the Open Compute Project is pushing designs from large cloud users such as Facebook into general availability.

The final strike at traditional storage hardware is the trend toward software-defined storage, which is aimed squarely at COTS hardware platforms. If we add all this together, it's clear that we are moving rapidly to a commoditization of storage, and subsequently a major lowering of prices.

With the demand side of storage set to explode over the next few years, it's not likely the total revenue of the storage market is going to shrink much, but I think we'll see a radical restructuring of market share, with new names entering the market.