Mad Dog 21/21: Seismically Active Storage
September 20, 2010 Hesh Weiner
During the early 1970s, the iconic large computer was the System/370-158 mainframe from IBM. It ran at and defined 1 MIPS. Its magnetic core memory technology was boosted by virtualization. But during the lifecycle of the 158, IBM fell behind.
Intel and others made solid-state memory that eclipsed core memory. Several companies made IBM-compatible disks, among them StorageTek, now part of Oracle. StorageTek, for awhile at least, became so good that IBM had to enlist it as a supplier. Now another earthquake is coming. This could be the big one. It could wipe out IBM’s remaining significance in enterprise storage.
For starters, storage drives are really quite a lot better than they were even a year or two ago.
Solid state drives are getting bigger and cheaper. And the software in storage controllers used to manage SSDs has become pretty good, too. All the major server operating systems recognize SSD technology, too, and include drives for the devices. (IBM has been supporting these drives on Power platforms running i 6.1 or later for more than a year.)
Spinning disk technology is shifting toward SAS interfaces and disks are shrinking from 3.5-inch to 2.5 inch form factors. The results: a dramatic reduction in the physical size of drives and arrays, a huge decrease in the power required (and heat generated) per terabyte of storage capacity, and an increase in the amount of data that can be tied to a read/write head. Drive thickness has come down, too, further boosting density.
As these components have become better they have also become a lot less expensive. That makes it hard for even the most entrenched suppliers to constrain the rate at which external disk subsystems and in-the-box arrays lose value. That, in turn, makes it difficult for vendors to sell storage capacity on the promise that it will pay off over the long haul. On the contrary, the rapid rate of change in storage technology and pricing suggests a new model for storage sales: systems that make it very easy for customers to move from old to new components.
Vendors generally talk as if it’s easy for customers to move ahead as data storage technology advances, but this is a virtue that is a lot easier to promise than to deliver. A disk subsystem that was unquestionably a good choice a few years ago because it took advantage of Fibre Channel drive connectivity would today be seen as a cul-de-sac. But a few years ago nothing looked like a safer bet. The industry had not yet overcome some of the problems that afflicted serial SCSI and SCSI-over-IP technologies. Today, SAS (plus enterprise-class SATA) drives seem to be the components with the brightest future, and iSCSI interconnection appears to have more promise than FC. (This issue is not settled; FC advocates, and there are plenty of them, say their technology, including drive-level interconnects, is the way to go, and then there are those who want converged networking, running Fibre Channel over lossless Ethernet.)
There simply isn’t the kind of vision in the storage arena that there was in communications when TCP/IP was formulated. Yet that is probably what IT needs to allow customers to invest with more confidence that their systems will be able to evolve without trauma. Basically, customers with IBM i, Unix, Linux, and Windows servers worry that their storage strategies could end up in the kind of swamp that has trapped IBM mainframe sites. IBM mainframes, even the brand new models just rolling off the Pokie assembly line, use storage subsystems that emulate 1980s-era model 3380 disks.
Look under the covers of the latest arrays from IBM, EMC, Hitachi, or Oracle and what will you find? Very exotic ways of presenting data organized as it was two decades ago to glass house systems that are not vastly different from the batch systems of twenty or thirty years ago. These batch systems are propped up by microcode or millicode or maybe millipedes on chips that are genetically modified offspring of Power Unix engines. The owners of mainframe shops are like deer in headlights. They know that one day soon some upstart competitor is going to move against them using arrays of X86 servers (or, if IBM is lucky, Power boxes) and banks of state-of-the-art disk arrays. The last holdouts are likely to be giant banks, corporate potentates of a business culture that are, under present conditions, too big to worry about anything. It’s going to be quite a shock when the iBank or Googlebank comes along.
The situation is quite dangerous for the storage specialists like EMC and NetApp as well as for the systems vendors serving enterprise accounts. External storage, which is the only kind that can be used with mainframes and which is also the kind of storage widely used with other kinds of large servers (or server farms), suffered a decline in revenue of nearly 9 percent during 2009, according to Gartner. IBM, with a revenue decline of less than 1 percent, was the leader, although its volume was a bit over half that of EMC. (The Gartner data is presented here.)
IBM’s outstanding performance seems to have been due to excellent results from its new XIV line of arrays, sold to shops with large ‘nix servers. The XIV devices that are much cheaper than IBM’s flagship DS8000 products because they take better advantage of current disk drives, but for marketing or technical or corporate political reasons (no outsider knows the correct story), XIV subsystems don’t have the FICON interfaces that big iron requires. Despite what IBM characterizes as its excellent market reception, XIV might soon be RIP; its inventor is reported to have left IBM.
Whatever is going on inside IBM’s storage business, the whole segment picked up speed during the first quarter of this year, growing by 17 percent according to IDC. The big gainer was EMC, which beat its first quarter 2009 results by an impressive 37 percent. IBM’s revenue grew 21 percent, which means that even with all the added business EMC increased its lead. When internal disk sales are added to external sales, the picture changes quite a bit. IBM ends up in third place after EMC and HP. HP is selling so much in-the-box storage that it nearly matched EMC’s revenue during the quarter. Dell sold a lot less in-the-box storage than HP but enough to just about catch up to IBM. If Dell has a strong second quarter, it could pass Big Blue in the storage business, a possibility that shows just how much competitive pressure IBM must contend with. (Our detailed analysts of IDC’s take on the total storage business during the first quarter is presented here.)
The race in storage for dominance in iSCSI or FC network attached storage is brutal right now. Hewlett-Packard and Dell were, at the end of August, practically murdering each other in a contest to acquire 3PAR, a contest won by HP, if you happen to think buying 3PAR at the $2.4 billion price HP paid is a win. 3PAR thinks it is XIV that works; IBM still thinks XIV is XIV that works.
The changes in storage technology, pricing, and vendor practices just won’t quit. Complacent users, shops that don’t add disk capacity very often, may think they are safe, but that is probably not the case. Chances are, they will be hit by friendly fire. They may discover this when they try to boost the storage on servers with in-the-box arrays and have a difficult time locating suitable disk drives unless they are willing to pay a very large premium to suppliers who know their older disk drives are vintage products and can command vintage prices.
When technology moves ahead very quickly–and that is the case with storage right now–it takes no prisoners and leaves no wounded behind. Corporate support groups that believe their companies will need to maintain or expand the storage in aging servers or older external arrays are starting to find that some components are awfully hard to source compared to, say, six months or a year ago. Some groups are defensively building up spare parts inventories while drives are still available. Others are developing game plans that involve the acquisition of refurbished disk drives, a practice that most of the time is confined to the tough and savvy techies who keep ISP and service bureau server farms running.
The disk drive business simply cannot afford to produce in small volumes, so when a market segment dwindles the big name suppliers simply phase out production. That is what is happening in the 3.5-inch drive market right now. If your company does some of its own maintenance and upgrading, and it depends on 3.5s, it’s time to put a few spares on the shelf. If you think you will be able to get 2.5s and adapter frames that will work in your servers or arrays, now is the time to test your beliefs. If you buy maintenance services, keep a keen eye on the costs; the price of insuring the survival of older storage equipment might go through the roof. If you are shopping for an external array, ask prospective vendors what they think is going to happen during the next year or two or three and how their products will fit in–or not fit in–as the world moves on.
And if you have to do bean counting as part of your IT management job, now might be a good time to adjust your spreadsheets to reflect the likelihood that storage bought this year and next year, if not forever from now on, is going to lose its value very quickly and run out of practical life much faster than the older machinery you already have on your floor.
Are you mad at somebody who uses computers? Tell them you hope they live in a time of great technical progress.