Some storage array technology bets are certs but others seem to deny the wisdom of the crowd. Often, the companies...
that bet on the outsider techs are large and successful, and sometimes it seems baffling that they go striding off in a direction no one else is following. But it's all down to the difference between well-understood technology on the one hand and what the market is doing on the other.
Take the areal density of the hard disk drive (HDD) in a storage array. This is rocket science, with the technology bringing together the extremes of physics and chemistry into mechanics and electronics, and with firmware to manage it.
This is a cert. The processes are so well understood, the technologies so fertile and predictable, that we know that 1 TB two-platter 2.5-inch hard drives are coming, with 4 TB four-platter 3.5-inch drives not too far behind. The only questions are who will announce and ship first -- Hitachi GST, Seagate, Toshiba or Western Digital -- and when? Will it be by the end of this year or early next?
Beyond that, we can see the next generation: 2 TB two-platter 2.5-inch and 8 TB four-platter 3.5-inch hard drives.
Solid-state drive (SSD) capacity increases are a little harder to predict, partly because the measure of SSD value is related to the HDD yardstick. How close to HDD's cost per gigabyte do SSDs have to come before mainstream markets will shift? And by mainstream I mean not just the high-end, slim-at-all-costs notebooks and performance-with-no-limits desktops.
That question is complicated by the write/erase cycle issue and is exacerbated by the needs of multi-level cell (MLC) flash technology. Is the road to SSD market takeoff via process shrinkage (Micron/Intel) or MLC technology (SanDisk/Toshiba)? No one knows yet, but process shrinkage doesn't mean you need a more sophisticated controller as the MLC route demands.
HP, Xiotech, 3PAR and NetApp diverge from storage array path
Now the trickier stuff: There is, for example, a vast market for the shared storage array, with suppliers pushing home the message that you need it to make a success of server virtualisation.
Except for Hewlett-Packard (HP), which claims the market is moving away from the storage array and toward direct-attached storage (DAS). A source at research firm IDC suggested offhand this might be because the EVA product line has "needs" and hasn't been as successful as HP might have hoped. On the other hand, this should be taken seriously, because this is HP talking and not some flaky wannabe supplier from San Wherever in the Silicon Valley.
HP's pitch is that customers are going to start buying racks that integrate virtualised servers and networking with storage local to these servers. And my IDC source supports this view of customers buying integrated IT pods, so HP could be right. But this means every other shared storage supplier is betting on the wrong horse. HP's stance is really counterintuitive.
Another outsider bet is the one being taken by Xiotech. The storage array of the future -- according to Xiotech -- will be one that just looks after reads and writes and ensures that data is stored reliably and is available. Naturally, Xiotech's ISE storage enclosure is the best technology for this. Free the storage array from bloated and complex controllers is the company's message, and leave upper-level storage operations to be carried out by server-level applications such as Symantec's storage suite or IBM's SAN Volume Controller (SVC).
Nearly every storage array supplier depends on the profit margins from the software added to the controller on what is basically a box containing commodity hard drives, interfaces and controller hardware. Xiotech is betting on its ISE drive technology and is looking for OEM contracts as well as selling it via the indirect channel. Winning the first OEM contract and then another will be a validation of a technology concept that has been struggling to succeed for many years.
3PAR is now betting that storage arrays will move from multiple disk tiers topped by SSDs to just two tiers: SSDs for performance and SATA for bulk capacity, with Fibre Channel and SAS drives squeezed out of the enterprise array business.
This is not a message that will be welcomed by Seagate and Hitachi GST. 3PAR is basing its strategy on cheaper Mach8IOPS SSDs, which are less than half the cost of the pioneering ZeusIOPS product used by EMC, IBM and others. It's a distinctive pitch and 3PAR might be right, with some vendors such as NetApp being quite sympathetic to this technology direction.
Meanwhile, NetApp is making its own counterintuitive bet. It says a dual-controller storage array with very clever software and an SSD controller can do most of the storage work customers need. NetApp's not going to bother with the markets for things like extreme scale-out filers (targeted by HP/Ibrix, Isilon and IBM SONAS) and deduplicating virtual tape libraries (served by Data Domain, Quantum, Sepaton and others). Instead it's going to stick to its Data Ontap/FAS knitting and bet that data going to virtual tape libraries today will go to the cloud tomorrow.
No other mainstream storage array vendor has adopted such a stance. It's as different from the mainstream as HP's focus on DAS. There are critics who say that NetApp would say this because Data Ontap is now so complex that extending it in new directions is becoming impossible. Sour grapes, says NetApp. We don't need multiple product lines as our competitors do because our technology is not limited in the same ways as theirs.
So here we have two networked storage industry giants, HP and NetApp, preaching revolutionary and unlikely messages that could be right, but could equally be horribly wrong. Smaller vendors like 3PAR and Xiotech also preach messages that could be very negative for other suppliers, and are individualistic to say the least.
Server virtualisation and the cloud shake up storage array supplier bets
The storage industry is in ferment. Server virtualisation and the cloud are delivering seismic shocks to an industry that had become complacent. If customers buy integrated IT stacks in racks, with server hypervisors being the point of administrative control, then how will network storage array suppliers develop and sell products that will succeed in this coming high-utilisation and hyper-integrated cloud rack world?
This provides the background to all the various networked storage array supplier bets. Cloud IT rack integration could kill several of today's successful networked storage companies; it could squeeze them out of the market if they have no server supplier tie-up and no reliable channel to sell product to cloud service providers.
Revolution is in the air, and it's not surprising that we're seeing revolutionary strategies come from previously complacent companies. Looked at in the perspective provided by these big questions, the issue of when 1 TB two-platter drives are coming is trivial, or relatively trivial.
There's no question such drives are needed, and the four HDD suppliers know this. But they've been through a tortuous supplier shakeout. They're the survivors, and what looks likely is that the upper levels of the storage stack could be set to go through a shakeout too -- with a consequent contraction of the number of suppliers and painful exits from the industry.
Chris Mellor is storage editor with The Register.