It's MIPS Again, at Least for Hardware Pricing

Published: June 13, 2006

by Hesh Wiener

Officially, as far as IBM is concerned, MIPS, short for Millions of Instructions Per Second, has been a forbidden acronym. IBMers may talk to each other about mainframe MIPS, and they might even use the word with customers, but for years it's all done under conditions that allow deniability. Now things are changing. In fact, IBM seems to be ready to do more than talk about MIPS. It's actually going to start selling MIPS, or so it seems. It's never possible to be absolutely sure in the murky world of mainframe marketing, but all the signs are there.

In a sense, IBM doesn't really have a choice. With the introduction of the System z9 BC line IBM is offering mainframes that run at 73 different performance levels using only two different multiprocessor configurations in one physical frame. All the performance differences are due to microcode (or, if you prefer, firmware) that governs the throughput of the machines. This concept allows IBM to make only two variations on one platform to serve all its customers but still offer lots of choice when it comes to performance and, of course, price. From a manufacturing perspective, it makes the whole plan viable even if IBM sells only, say, a thousand or fifteen hundred boxes during the production life of the system.

To make matters a little more complicated, IBM's mainframe business has had to sever the pricing of software, based to a large extent on performance, from the pricing of hardware, also based on performance. The software pricing is based on units of capacity called MSU, or Millions of Service Units, but the MSU is no longer defined in any technical terms, as it was at first. When IBM brought out the z890 line, it rated the machines at MSU levels that gave customers about 10 percent more performance per MSU compared to the z800. Another similar amount of MSU inflation is part of the ratings of the z9 BC compared to the z890 line it succeeds. These reductions in software charges, at least for packages priced base on MSU ratings, spur users to move ahead.

Hardware pricing can't change quite the same way. Mainframe hardware can't be sold if price/performance improves at a modest 10 percent rate every year or so when a new generation is put into production. Alternatives in the form of Windows, Linux, and Unix boxes provide price/performance improvements at a much faster pace (about a dozen times that rate, in fact, according to Moore's Law). That is one reason these machines have taken the center stage once occupied by mainframes, which are still only dominant, if under threat, at the largest sites and at midsize organizations that, as a practical matter, cannot replace their legacy applications with software that runs on mainstream platforms.

IBM's mainframe marketeers need more freedom to adjust pricing if they want to hold the low end of the installed base, where z9 BC systems and, perhaps, the newly announced z9 EC models that offer reduced capacity at reduced cost compared to prior z9 models are aimed.

IBM's z9 BC solution is ingenious, because it allows IBM to adjust the processing power to almost any imaginable level from 26 MIPS to 1,790 MIPS, according to my estimates. IBM's speed governing technology affects only general purpose engines. They can be throttled back while system management processors, IFL dedicated Linux engines, and zIIP or zAAP co-processors run at full speed. Memory and I/O run full tilt, too, and have essentially the same capacity across the product line; this has to be the case or bottlenecks would cripple the coprocessors that are so important.

It's a concept that may have originated with the green movement. People who want to save water put objects in their toilet cisterns to reduce cistern capacity. They don't change the size of the pipes linked to the appliance, or any other technical parameters. The result is a toilet with the same storage capacity and same I/O, but if you give it a big load you will see that the throughput is less. And that's the way it is with IBM's new mainframes, with one un-ecological exception.

A one drawback of IBM's MIPS management scheme is that systems with lower capacity use about as much electricity and generate about as much heat as ones with larger capacity. But this disadvantage is a characterization that only applies to general purpose processors. Because all the z9 BC's other processors run at full speed regardless of the mainframe MIPS provided by the box, the issues surrounding the energy efficiency of a machine is only a significant drawback to the extent the computer is dominated by general purpose MIPS-based engines. Basically, the higher the MIPS rating of a z9 BC, the greater the energy efficiency.

Whether they are paragons of efficiency or not, the z9 BC models ought to make it easy for IBM to give customers a system that precisely fits workload requirements.

Users who want more throughput and don't want to change anything else on their machines can pay IBM to change a machine's effective processing power by refreshing the microcode. And the measure of the processor capacity users buy will be the formerly abandoned MIPS. Users can also buy MIPS for temporary use under IBM's on-demand plans. The price of a MIPS, according to various sources, is about $1,600. But this pricing currently comes into play mainly when capacity is added on a temporary basis, and the cost is the price for 90 days' use of the MIPS. If the MIPS are to be used for 30 days, then the cost would be a third of that but at the end of 30 days the MIPS vanish.

Model changes done on a permanent basis can be less costly than MIPS on a temporary basis, but the pricing plan seems to be evolving into one that could dispense with specific models or speeds and instead offer mainframes of a particular class at any MIPS level the customer wants. The combination of continuously variable processor MIPS and increased virtualization may allow IBM, in the near future, to offer customers precisely the amount of computing power they need, to vary this capacity with changes in business conditions, and to give customers systems management software that will enable management of the capacity and cost of mainframe power.

Just as users will increasingly be able to adjust their mainframes, IBM has gained more freedom to adjust what it charges. IBM could combine the increased configuration flexibility of its mainframes with some more willingness to adjust the pricing of MIPS, IFLs, zIIPs, zAAPs, and other components as market conditions dictate. As a mainframe generation ages and competitive platforms pose a greater threat, IBM could adapt its price/performance model more readily than it could in the past. This could make the machines a lot more enticing to some customers who can't quite afford brand new mainframes but don't really want to be a generation or two behind the state of the art.

Even now, the new variable speed System z architecture is getting quite a lot of attention from customers who formerly hadn't found the mainframe quite as interesting. It's hardly a wonder, then, that these days IBM's System z division is flushed with pride.

RELATED RESOURCE

System z9 BC MIPS Table