Mad Dog 21/21: The Next (And Maybe Last) Mainframe
February 13, 2017 Hesh Wiener
For the past several mainframe generations IBM has announced new large systems about every three years. The current line of big iron, the System z13, emerged in 2015, so it wouldn’t surprise anyone if the z14 or whatever it is called reaches customers next year, in 2018. Meanwhile, this year Ginni Rometty turns 60, IBM’s traditional CEO retirement age, and, like her predecessor, she could stick around one more year as the company’s chairman.
The upshot: Rometty’s successor as CEO will enjoy the sales boost of a fresh mainframe product cycle. And if IBM is making money in mainframes, that takes some of the pressure off Power Systems.
When a new mainframe cycle begins, IBM sells a lot more than computer hardware. It also issues updated systems software, rolls out enhanced middleware, and freshens up its menu of support services. It gets users to go for the while package by offering attractively priced bundles that include financing. Typical IBM mainframe deals run for three years, during which time pretty much every item in the bundle is prices lower than its nominal list price if sold separately.
At the same time, IBM’s deals on prior systems are coming to an end. These deals usually cannot be extended at the same timers. Instead, the software and support parts of the deal rise in price. Hardware monthly lease costs may remain the same, but even so the total outlay for the old package rises to a level that’s higher than the cost of a new deal with new equipment and a path to updated versions of key software products. Forward migration, for mainstream mainframe shops, is usually the most practical course.
Because customers know the new systems and new deals are likely to arrive soon, they are reluctant to boost the capacity of installed systems even if their workloads are on the rise. The upgrades feel quite costly when it is late in a mainframe’s new product cycle. To encourage acquisitions by growing shops, IBM may reduce prices or offer extensions of the plans that keep software and services costs under control. Or they may offer upgrades under terms that allow the new configurations to be replaced at the same date the original base machine was slated for a fresh deal. It’s very easy for IBM to do this, because all the hardware for upgrades is already in mainframes sitting on customers’ glass house floors. All IBM has to do is turn on the additional capacity, which is done by installing altered firmware, and adjust any other configuration details that might need revision. The extras might include adding (or activating already installed) memory or provisioning more channels.
Upgrades late in a mainframe’s product cycle provide IBM with excellent profit margins even as they may give users a chance to get extra processing power at bargain prices.
These deals at big iron shops are likely to become very important to IBM this year, giving the company a chance to boost sales, bring a lot of money to the bottom line, and help the company justify a big bonus and grand goodbye to Rometty as she wraps up her tenure.
The nature of IBM’s mainframe strategy makes for a unique manufacturing plan. Basically, once IBM gets all the kinks out of a new mainframe system it just goes ahead and builds all the components it needs for the entire product cycle as quickly as it can . . . as long as the pace allows for good cost control. By the time the first units of a new mainframe generation start reaching customers, production is well underway and some components, such as processor chips, may all be done. IBM probably makes all the chips for a whole generation of mainframes in one batch, a process that may take weeks or months to complete but which is only done once per generation.
By now, IBM may be well along the way making chips for its 2018 mainframe models. It is also likely to be nearly done finalizing the subassemblies that hold the chips, components that used to be called books but which may get new names as IBM fabrication methods evolve.
This is quite different from the way X86 servers are made. Chips for these machines are made in many batches during a particular chip generation. There may be process refinements or other small updates along the way. But basically, Intel and other CPU chip makers serving high volume markets spread their circuit production across many batches and several wafer fabs during a production generation that may last several months, perhaps a year or more in some cases.
IBM’s Power Systems have some manufacturing techniques in common with mainframes and some with X86 servers. There’s enough demand for Power chips to justify multiple batches during a generation that might last three years, which is now about the same time as a mainframe chip generation but more than the production lifetime of an X86 server chip. (IBM often tweaks a Power chip with a so-called plus model, often with a process shrink, but it did not do that with the Power8 chip.)
IBM mainframes have longer lives inside IBM than they do in users’ glass houses. While IBM likes to churn its installed base using three-year deals that make for a marketing generation that may run up to five years (because not every user moves to new iron during its first year of availability), IBM’s service groups including those that provide IBM big iron cloud services may keep equipment quite a bit longer. Lots of IBM services customers stick with software for five to ten years, maybe longer. If their applications don’t need the features provided by the latest versions of various systems, middleware, and applications packages, there’s no reason for them to change. IBM’s services groups like to stretch the life of mainframes and their software. They are not subject to the deals users get when they install mainframes under three-year terms that end with a price boost. So machines IBM sells to its services groups, including systems taken back from end users as these customers replace installed equipment with newer processors, live on, bringing in revenue and working as flawlessly and tirelessly inside IBM as they do during their youth in users’ glass houses.
The economics of the services business and the various methods used to keep mainframes running far longer than they are in new production are well known outside IBM. IBM’s services rivals do pretty much what IBM does to make mainframes yield revenue and profit for several years longer than they would in a typical user’s shop.
Users can keep their mainframes for a very long time if they wish, and some do, but often this is because these customers want to freeze the technology in collection of legacy applications and are even willing to pay more for old hardware, it software and its maintenance than they would for newer machines, because the newer machine’s may not support old code.
This whole process begs the question: What if IBM stopped making new machines that were different from old ones? Sure, IBM might want to add new functionality or change the balance of internal processing capabilities as the market’s needs evolve. But it might be possible for IBM to do all this with firmware and thereby keep 2018’s mainframe hardware viable in users’ sites for ten years or longer rather than the three to five years of current and recent mainframes.
IBM could do this by making much more hardware for its next generation of mainframes, perhaps two or three times as many chips, which would not cost proportionally more to fabricate, make more subassemblies and the components that they include, and plan a combination of firmware, systems software, middleware, and applications upgrades that still give customers progress.
The result of this change in strategy would be to get IBM mainframe hardware out of a race against other processor chips that it can only win by investing an inordinate amount of money and technical talent. That same talent and probably a lot less money might yield more progress applicable to end user’s requirements than a continued effort to build snazzier processors in ever-decreasing box volume (even if CPU chip counts keep rising).
IBM probably doesn’t have this option with Power systems. Now that the Power world seems to be turning into a Linux engine market, IBM cannot slow let alone freeze its Power architecture and seek performance and functionality progress via the means that remain available to its mainframe group. Power has to go head-to-head against X86 boxes and, soon, ARM machines, and maybe Nvidia and AMD engines, too.
So, outside the mainframe world, IBM probably has to keep competing the way other computer companies do, except possibly for IBM i and AIX customers running on Power Systems. But in mainframe country, IBM has much control over the big iron environment that different rules govern IBM and its customers.