The enterprise SSD market is maturing, and, with maturity, comes a rash of acquisitions and structural changes to the business.
Intel has been the market leader in enterprise SSD for a long while. Its x25 product was the first drive properly targeted at that market, and the company has continued to build on the product.
Not unexpectedly, the giant of SSD takes second place. Samsung is the clear leader in the overall SSD market, with over a third of the business. They’ve overcome a lack of knowledge of the storage industry to be the major player in the game.
There have been a couple of recent large acquisitions in the space. Western Digital is buying sTec, ostensibly for its technology, though hanging on to its market share would spring WD to second place in the enterprise stakes. SanDisk is acquiring Smart Storage, which is more synergistic than the sTec deal, and that will bring them into contention for third place.
The game is still dynamic, even though, at five years old, the market is starting to mature and move into a consolidation phase. PCIe is a major alternative in the enterprise, and FusionIO owns the lion’s share, despite leadership issues and falling prices.
But Micron Technology and WD's HGST division are eyeing a drive form factor alternative to the PCIe card, which will give users a more comfortable use case of removability, without sacrificing performance over the PCIe accelerator cards. These PCIe drives will get motherboard support in the next few months, and could be a serious alternative to the card form factor. FusionIO had better watch out!
The market overall is looking mature, with the established players in control, and the wannabees looking for a safe haven. Another year or so should see the small fry exiting the game, either by acquisition for IP or just lack of a viable market.
Intel appears to have a strong sustainable position. Its ties to Micron, and its ability via reference designs and chipsets to control the availability of motherboard ports for PCIe drives and NVMe support, gives the company an edge.
Samsung is likely to maintain its spot in the number two position. The company is building on its technologies, and has a strong edge from the flash supply side of the market. It would be a good guess that Samsung has the cheapest flash chips in the market, just due to volume.
The WD and sTec combination will grow just a bit in total share. sTec’s products were declining fast, so translating market share is more difficult. The confusion that comes from trying to join the pieces of companies together will be tough, especially as sTec comes so soon after the HGST acquisition itself.
In the PCIe stakes, the pressure on FusionIO will come from the OEM market. Companies like EMC are still buying product, but some of the big players, like NetApp and Google, are already rolling their own. While Google’s products aren’t likely to be resold, EMC could follow the same model, pick up the controller and code from the market, and have an ODM make boards. Or, they could buy FusionIO!
All-flash arrays add a third product family to the enterprise equation. These boxes are like super-drives in structure, with lots of individual flash drive modules tied to switch and control structures. An array of PCIe drives might describe them, though the internal interfaces can be any of several standards.
The all-flash array market is truly “emerging.” Violin is the current leader, ahead of a second tier of Big Iron vendors such as EMC and IBM, and a third tier of startups. This isn’t the whole story, since there are OEM deals that mask the total production of the startups. Of course, it’s anyone’s guess who’ll win out in the game, but the likely path is for the Tier 2 companies to absorb the Tier 3 startups.
Now, all of this activity might become moot in a year or two. Intel and Micron are part of a consortium to create a 3D memory. The Hybrid Memory Cube (HMC) adds tightly-coupled memory stacks to CPU modules, boosting performance into the terabytes/second range. The model specifically provides a vehicle for tightly coupling flash devices to the architecture, and uses a memory controller architecture that is asynchronous and highly parallel to allow for a mix of slow and fast devices.
HMC could hit the PCIe space like a photon torpedo, moving the accelerator on-CPU-complex and into Intel’s close control. Perversely, and this depends on pricing and availability of NVMe, it might boost the PCIe drive form factor business, due to the IO bandwidth needed to feed the CPU beast. This issue will be much clearer once the HMC units begin serious volume shipments in 2014.