Analysis Mainframe maker IBM and its few remaining peers that sell their own mainframes are probably cheered by two recent surveys of mainframe shops, one that came out of market researcher IDC and the other out of mainframe software maker CA. But maybe not.
IDC polled some 300 IBM mainframe shops and kicked out a study with the ponderous name Directions in the Multi-Platform Data Center 2009-2013: Today's Workloads and Future Outlook. This is IDC document 219797 and the 64-page report packs a mainframe-class price, at $20,000.
Given the value that IDC places on this report, the tidbits that it leaks out to the IT trade press are pretty thin. But the box counter was willing to reveal that nearly half of the 300 respondents in the survey said they would increase their annual spending on mainframe hardware and software. And in another comment, IDC said that mainframe shops "can plan another wave of investments in the System z platform over the next 2 to 5 years, given the system's high availability, reliability, and security for mission-critical applications."
This is the same old mantra that IBM and other interested mainframe parties - such as IT consultancies and third party software suppliers that peddle products for mainframes - have been chanting for decades. They don't get tired of saying it or hearing it.
What I want to know is if nearly half of the respondents said they were going to increase mainframe budgets, what were the more than half of mainframe shops planning? Do they expect their budgets to be flat? Are they planning cuts? And if some shops are planning cuts - which seems very likely given the intense pressure on all platforms to offer better bang for the buck - will these cuts be so deep that they wash out any planned increases by other shops? The IDC report, or more precisely the summary that the public can see, raises more questions than it answers.
What can be said is that by corralling some mainframe engines, slashing their prices, and letting them run Linux instances or speed up Java or DB2 workloads, IBM has been able to keep the installed base of mainframe processing power, as measured in MIPS, growing - even as it has its ups and downs each quarter. And the downs have been especially sharp since the economic meltdown kicked in a year ago.
The advent of lower-priced System z Business Class machines, a sort of midrange mainframe akin to the ones that acted like feeder boxes into bigger iron in days gone by, has also helped keep mainframe shops from dumping their big iron for Unix, Windows, or Linux iron by reducing the hardware and software costs of mainframes.
Mainframe deals take a long time to get done, and to stop. IBM's System z sales were up 25 per cent, while MIPS shipped rose an impressive 49 per cent in the third quarter of 2008, before the economy started to collapse in the United States.
In the fourth quarter, mainframe sales dropped six per cent, while MIPS shipments and MIPS grew by only 12 per cent. By the first quarter of 2009, System z sales fell 19 per cent.
But because IBM was peddling the System z BC and aggressively pushing its Integrated Facility for Linux and zIIP and zAAP accelerator engines (which cost a fraction of a mainframe engine set up to run the z/OS operating system), the company was still able to crank up the MIPS by 18 per cent. (I get the impression that IBM's System z managers are getting bonuses based on MIPS and profits, not revenues.) In the second quarter of this year, against a very tough compare, mainframe hardware sales declined by 39 per cent and MIPS shipped fell 20 per cent.
It came as no surprise, then, in August when IBM launched System z solution edition bundles, which are designed to only have a 20 per cent premium for a bunch of different workloads compared to Hewlett-Packard Integrity servers running HP-UX, and slashed prices on specialty mainframe engines by half.
We're about to see how elastic demand is for mainframe processing capacity. So far, IBM has done a pretty fair job of keeping hardware sales relatively flat on an annual basis. My own estimates, based on the thin data that IBM gives out each quarter, peg mainframe hardware sales at $3.61bn in 2006, $3.18bn in 2007, and $3.61bn again in 2008. The first half of 2009 has been particularly awful, with sales down 28.4 per cent to $1.18bn, by my estimates. IBM has to rake in $2.43bn in mainframe revenues in the second half of 2009 to break even with 2008, and $2bn to match the not-so-great 2007 year.
Why IBM is still excited about the mainframe
I think it is highly unlikely that mainframe sales break $3.5bn a year very often, and I doubt they ever see $4bn a year again unless something really weird happens in the systems business. IBM's Power-based server business, which includes AIX, Linux, and OS/400 machines, accounted for around $5bn a year in sales in the past three years, and have held sales in the first half of 2009, at $2.2bn by my estimates.
Only when you throw in the substantial - and wickedly profitable - mainframe software stack into the mainframe sales equation can you understand why IBM is still excited by the mainframe. This software is rented on a monthly basis for the 10,000 mainframes installed worldwide.
For most users, there is not a competitive product they can buy. And this is where IBM gets the billions and billions of dollars to do share buybacks. IBM could give away mainframe hardware and probably still have the same profit margin it has with Unix machines. Legacy code is a sweet, sweet thing for IBM and for customers who just can't imagine recoding their mainframe apps. So long as they have money to cut those monthly checks.
The IDC statement on the mainframe report didn't say much else, other than IBM's use of specialty engines has revitalized the platform, that processing power and uptime were key differentiators for the mainframe, and that saving money was the number one reason for moving applications off the mainframe.
And now to CA, the software maker, which is deeply involved in the mainframe game. The company surveyed 100 mainframe shops in July and August and was willing to share more details about its findings than IDC was. CA talked to companies that ranged in size from $500m to more than $10bn in annual sales about their mainframes and what they are happy with and worried about.
Among those mainframe shops surveyed, 95 of the 100 said they had distributed servers (Windows, Linux, and Unix) in their data centers as well; 26 said they outsourced applications, 19 said they used various software-as-a-service officers, and 13 said they had other kinds of mainframes.
To get an idea of how deep these companies were into mainframes, CA found that a quarter of the mainframe shops had an aggregate of 20,000 MIPS or more in their data centers; another 10 per cent of the shops had between 10,000 and 20,000 MIPS and 22 per cent had between 1,000 and 10,000 MIPS. Nearly another quarter had relatively modest mainframe setups, with under 500 MIPS, and only eight per cent said they had between 500 and 1,000 MIPS.
About one-eighth of the shops said they didn't know. If you do the weighted average on the midpoint of those shops, you're talking about an average shop having around 7,650 MIPS, and at around $1,000 per MIPS for the iron (taking into account a mix of regular and specialty engines), call it somewhere around $8m in mainframe iron. Those customers polled probably have somewhere around $675m in mainframe hardware.
In aggregate, those companies polled by CA allocated about 38 per cent of their IT budgets to mainframe hardware and software, while 56 per cent of the budget covered those distributed platforms; the remaining six per cent of the budget went to hosted apps of one sort or another.
When looking at new workloads, mainframes are beat out by distributed platforms except when it comes to transaction processing applications. But a little more than a third of the time, no matter what kind of application is going into the mainframe shop, the mainframe gets a chance.
Attesting to the backbone nature of the databases running on mainframes at mainframe shops, a third of the companies polled said that if the plugs were pulled from their mainframes all their applications - even those running on distributed platforms - would fail. Another 27 per cent said a lot of important applications on distributed systems would fail, but some could keep running; 24 per cent said some would fail and another 13 per cent said there would little impact.
The mainframe has its place, and like every other platform, it has to compete on the merits. Quarter after quarter, year after year. ®