|HOME||PUBLIC LIBRARY||ANOTHER PERSPECTIVE||INFOPERSPECTIVES||CONTACT|
Now that the z9 BC mainframes are making their way to the market, the z800 has reached the terrible two. Mainframes that are two generations old can drop from sight, or at least viability, like boat anchors. Prices of used z800s have plummeted, right according to the traditional script. But this time things might turn out differently. Depending largely on IBM's practices, the z800 could be reincarnated, not as popular platform for general purpose use as a mainframe, but as a cost-effective host for a diverse collection of workloads.
When it comes to running a stack of IBM software packages and the mainframe applications they support, there's no question the z800 is a goner. IBM's software charges are based on MSU performance, and under IBM's capricious MSU rating system the newer z890 gives about ten percent more performance per MSU than the z800. The z9 BC ratings give the brand new models another 10 percent. In simple cases, the software for a z9 BC will run 19 percent faster less than the same stack on a z800 with the same MSU rating, but at the same cost. That translates into 19 percent workload growth on IBM's nickel before a customer has to move to a higher, more costly machine with a higher MSU number.
A z800 doesn't have the finer granularity of a z890, let alone the 54 performance choices offered by the z9 BC. So, when a z800 user needs a hardware upgrade, the software license fees speak for themselves. The case for a new machine compared to that upgrade will almost certainly include an attractive if not compelling set of figures reflecting the software cost advantage of the z9 or z890. The technical advantages of a newer machine, such as the availability (and lower price) of zIIP, zAAP, and IFL Linux engines, only strengthen the case for forward migration.
But that's not the whole story, not by a long shot. With the smallest used z800 machines available for less than $50,000, and the largest four-engine boxes offered for about $175,000, imaginative customers may find the machines just too cheap to ignore. For some, the purchase price can be the full price, because there may be no reason to pay for maintenance on a z800. Getting a spare or replacement processor can be more affordable than paying for IBM repair services, particularly if that spare can be put to work until (or unless) it's actually needed.
A cheap z800 running z/VM and Linux can provide a single box alternative to a Linux (or Unix) server farm, and some shops will see right away that the mainframe alternative is preferable to racks of x86 boxes and their networking apparatus. Another option comes from IBM in the form of z/OS.e, a runtime subset of z/OS that can enable users to offload costly capacity from a newer, general-purpose mainframe that will still be needed for work that can't be performed under z/OS.e. The z/OS.e choice is, of course, irrelevant to VM/VSE shops.
Lastly, a cheap z800 can provide a disaster recover environment, very possibly at lower cost than a DR program bought from IBM or one of the other big DR service firms. There's no license charge as long as the emergency software isn't in emergency use, and as we've pointed out, the cost for maintenance is a matter entirely in the users' hands. Nervous users who don't want to pay for maintenance on a DR box can fire the machine up as often as it takes to prove the backup iron is in working order. The numbers tell the story: a whole used 2066-001 costs $50,000 to $60,000; IBM maintenance on the box is about $50,000 a year.
The one catch for users who seriously contemplate getting a used z800 will be finding a box with an acceptable configuration. While whole machines are cheap, upgrades are not. That means a user might be able to afford a used z800 but not the cost of getting all its memory enabled, if the machine a buyer locates happens to be a little short on main storage. The same goes for channels. Because extra memory and channels don't add to maintenance charges (if the user actually wants maintenance), there's no penalty for running a fatter configuration. That means, in a market where the supply of used machines can be thin, fully configured boxes are a lot more attractive than skinny ones. Even users who only need lean configurations would be wise to get the most memory and channels they can, because the extra features are going to add considerably to any salvage value that z800 might have when it's sold or traded in. The extras are certainly not going to diminish the machine's value.
The main source (and sometimes the only source) of a used z800 is IBM itself, which sold the bulk of the base under leasing arrangements. The z800 started reaching users in early 2002, making the machines as much as four years old. IBM pretty much stopped selling z800s in 2004, when z890s became available, although it did provide z800s to users who for financial or other reasons wanted them for some time after z890 shipments began. Typical IBM leases run for three years, so many z800 financing arrangements have already run to term and the remainder will terminate by the end of next year (with a few possible exceptions due to extended leases and other unusual occurrences).
Still, just because IBM has barns full of old z800s there's no assurance Big Blue will be selling them off. If IBM put all its used z800s on the market, supply would undoubtedly overwhelm demand, and the amount IBM could get for any of its secondhand machines, if it wanted to clear out its warehouses, would be even lower than today's bargain prices. Even if you factor in demand that focuses on heavily configured boxes, IBM could easily drown the market. Boosting memory in boxes that are loaded with hardware is just a matter of microcode. The pool of used channel cards in IBM's hands far exceeds the number that the market could absorb. For IBM, adding channels or memory to a z800 is hardly rocket science, and certainly not very costly.
It is, however, in IBM's interest to get as many customers as possible to buy new hardware rather than used, particularly if Big Blue can still get some value out of its collection of old z800s by using them in its services division. In other words, IBM might not be too keen on selling z800s on the cheap if they in some way cut into sales of new z9s or the remarking of z890s that come back to its finance group.
IBM can try to encourage customers to only use z800s for jobs that would otherwise be run on alternative platforms, or which wouldn't be performed at all, but it can at best only influence this part of the market and it cannot completely control its user base. That makes any campaign to peddle used z800s a bit of a gamble, something IBM is often not inclined to do. And the chances that IBM would do something really creative, such as find a big Internet portal or search or sales outfit or ISP that uses Linux and sell it dozens of z800s to replace hundreds of X86 or RISC boxes, are slim to none. (There's also a real possibility that z800 I/O isn't up to the job compared to the aggregate disk transfer capability of the installed server farms.)
Despite any constraints IBM might place on the used z800 market, despite the relatively small base of creative users who have faith in the mainframe, and despite the very small number of used equipment dealers who have the skills and experience required to help a customer take advantage of bargain basement z800s, it is too soon to write off IBM's most affordable 64-bit big iron. There just might be enough users who see a z800 not as a dead dinosaur but rather as a way to make some nice shoes and handbags. If that's the case, the z800 might, in a limited way, make a fashion comeback.
— Hesh Wiener May 2006