A Radical Idea For IBM i Software Pricing

November 14, 2011 Timothy Prickett Morgan

Lately, I have been thinking a lot about IBM‘s pricing for the IBM i operating system and database. With Big Blue having long since converged its Software Group and its Systems and Technology Group, you would think that the heat would be off in some ways to upgrade hardware and for the company to focus on getting customers current on software. But IBM likes to sell new systems as well as operating systems to customers, and it needs to make money as much as midrange shops don’t like to spend it.

I decided to take a look at OS/400, i5/OS, and IBM i pricing over time, and like you, I remember the past being a lot more simple than it actually was. In the 23 years that the AS/400 and its progeny has been around, the company started out with a complete software stack that was priced based on many software tiers, roughly correlated to the performance and storage capacity of the system, and ranging from around 25 to 30 percent of the cost of a base system plus the base OS. In 1994, IBM made a bold move, shifting to five software tiers with much lower prices and adding on per-user charges in an effort to better compete against Unix and Windows systems. (This was the same year IBM moved storage under the skins and got away from external storage arrays.) For a lot of complicated reasons, IBM was not happy with this user-based pricing, and eventually it was spiked in favor of charging for 5250 capacity and treating it like hardware when we all knew it was software. At this time, in the late 1990s, the actual convergence of the AS/400 and RS/6000 Unix server lines was well under way, and Big Blue absolutely milked the OS/400 customer base, charging huge premiums for AS/400 and then iSeries hardware to pay for the competitive Unix wins upper management believed it needed against Hewlett-Packard and Sun Microsystems, which is now part of Oracle.

In the long run, Ginni Rometty, soon to be president and CEO of IBM, may be wishing that Big Blue had made 275,000 AS/400 customers happy instead of sacrificing more than half of them to win the Unix Wars. IBM made lots of money, but it didn’t keep its staunchest customers, and now many of those AS/400 shops are Windows or Unix shops and the Power Systems-AIX shops are by and large Oracle database shops.

In any event, in the past decade, IBM has tried a combination of user-based pricing and tiered editions of the OS/400 platform (in its many different versions and names) to try to make the system software more competitive against Unix, Windows, and now Linux platforms. And the way it breaks now, the low-end machines have a mix of per-core and user pricing while the high-end machines have per-core licensing only. (I gave myself a headache trying to find all the lost announcement letters and analysis I have done in the past two decades covering all of these price changes.)

The hardware-software pricing pendulum has swung back and forth over the years, as you can see from the tables below. The important thing is that we are in an era when the software component of a base machine–a chassis, the processors, base memory and disk, and operating system–is considerably more expensive than the base processors that support it. And perhaps more significantly, we are talking about an IBM i customer base that is much less broad, in terms of spanning the complete Power Systems product line. IBM is giving most customers far much more iron than they need for their modest workloads. And because they are not buying expensive hardware, IBM is trying to make it up in software fees. This, of course, can set up a negative feedback loop and limit growth. However, at the same time, IBM’s top brass in the Power Systems division have made it clear to me that they are encouraged that sales of the IBM i software stack are on the rise.

Here’s what a 20-class machine looks like over time; the pricing and packaging of all of those machines was just too much to wade through in the middle 2000s. IBM was trying a bunch of different things and I suspect the packaging worked, but the machinery doesn’t fit into the simplified model I have here. Take a look at this table:

Cost Of OS/400, i5/OS, and IBM i on Various 20-Class Machines
		Operating	Base	HW + OS	CPW	SW	$/ CPW	$/ CPW	OS %
Year	Model	System	Hardware	System	Perf	Group	OS	HW	Of Total
1988	B20	$12,120	$37,920	$50,040	5.1	E5	$2,376	$7,435	24.2%
1990	C20	$13,000	$37,920	$50,920	5.3	E6	$2,453	$7,155	25.5%
1991	D20	$13,000	$28,500	$41,500	6.8	E6	$1,912	$4,191	31.3%
1992	E20	$13,650	$33,000	$46,650	9.7	E6	$1,407	$3,402	29.3%
1993	F20	$14,330	$37,000	$51,330	11.6	E6	$1,235	$3,190	27.9%
1994	300-2040	$19,600	$32,500	$52,100	11.6	–	$1,690	$2,802	37.6%
1995	500-2140	$34,000	$38,000	$72,000	21.4	–	$1,589	$1,776	47.2%
1997	600-2135	$13,000	$45,000	$58,000	45	P10	$289	$1,000	22.4%
1999	170-2408	$9,000	$22,500	$31,500	120	P10	$75	$188	28.6%
2001	270-2454	$9,000	$21,500	$30,500	240	P10	$38	$90	29.5%
2003	810-2465	$41,000	$9,000	$50,000	750	P10	$55	$12	82.0%
Time Passes, And IBM Bundles and Dices And Slices OS/400 and I5/OS In A Zillion Ways
2007	i5 520	$5,995	$41,505	$47,500	1,200	P10	$5.00	$34.59	12.6%
2010	Power 720	$7,980	$6,835	$14,815	23,800	P10	$0.34	$0.29	53.9%
2010	Power 720	$87,960	$16,995	$104,955	46,300	P10	$1.90	$0.37	83.8%

First of all, it is really hard to make a historical line that stretches back from the Power 720 in today’s lineup all the way back to the AS/400 Model B20 launched in June 1988. What is also obvious is that the amount of computing capacity in machines in roughly the same place in the AS/400-Power Systems lineup has grown a lot faster than workloads have at most IBM midrange shops in the past two decades. This was not, I believe, a foregone conclusion, but rather the result of the egregious pricing for AS/400 and iSeries hardware in the late 1990s and early 2000s that more or less drove AS/400 shops that would have much-preferred to keep their Web infrastructure and application servers on their IBM midrange platforms but were driven into the loving arms or gaping maw of Microsoft.

That entry Power 720 server, which has four Power7 cores running at 3 GHz, is arguably the best bargain that any AS/400 customer has seen in years. (That is not the cost of a fully configured system, just the processors, processor activations, and base OS on the four cores.) Look at how much more expensive it is to move to the eight-core processor. The operating system is more than 10 times as expensive for a machine that has twice the oomph. This is absolutely ridiculous, of course. It is enough to make you want to cluster four of these baby Power 720s together instead of buying up the IBM line.

Here’s how OS/400, i5/OS, and IBM i pricing stacks up over the years for 70-class machines:

Cost Of OS/400, i5/OS, and IBM i on Various 70-Class Machines
		Operating	Base	HW + OS	CPW	SW	$/ CPW	$/ CPW	OS %
Year	Model	System	Hardware	System	Perf	Group	OS	HW	Of Total
1989	B70	$75,900	$332,050	$407,950	20.0	K5	$3,795	$16,603	18.6%
	C70	No Such Animal, HA HA
1991	D70	$95,000	$310,000	$405,000	32.3	K6	$2,941	$9,598	23.5%
1992	E70	$99,750	$335,000	$434,750	39.2	K6	$2,545	$8,546	22.9%
1993	F70	$104,700	$400,000	$504,700	57.0	K6	$1,837	$7,018	20.7%
1994	320-2050	$128,600	$273,500	$402,100	67.5	–	$1,905	$4,052	32.0%
1995	510-2144	$89,600	$195,000	$284,600	111.5	–	$804	$1,749	31.5%
1997	620-2181	$77,000	$115,000	$192,000	210	P30	$367	$548	40.1%
1999	720-2063	$70,000	$225,000	$295,000	240	P30	$292	$938	23.7%
2000	830-2400	$70,000	$342,000	$412,000	240	P30	$292	$1,425	17.0%
2003	825-2473	$260,000	$70,000	$330,000	3,600	P30	$72	$19	78.8%
Time Passes, And IBM Bundles and Dices And Slices OS/400 and I5/OS In A Zillion Ways
2007	i5 570	$106,000	$172,000	$278,000	8,400	P30	$12.62	$20.48	38.1%
2010	Power 770	$636,000	$85,825	$721,825	77,000	P30	$8.26	$1.11	88.1%

Once again, you can see that relatively speaking, the IBM i software licenses on a machine represent a very big portion of the base system cost. To one way of thinking about this, the hardware has become relatively inexpensive. To another way of looking at it, IBM is charging a slight premium for hardware and a very high premium for software relative to the Windows alternatives at most IBM i shops. IBM is pricing the machine like a Unix-Oracle server in a world that is Windows.

As I explained in last week’s issue, if you assume low CPU utilization on the entry machines, median CPU utilization on midrange machines, and high CPU utilization on large machines, then the difference in the cost per CPW for an IBM i license on any of the Power7-based machines is not very large–except on that four-core Power 720, which obviously has a very low price designed to be appealing to the vast majority of IBM i shops.

While this low price is good, what I see when I look at this is that IBM has essentially conceded that it cannot attract workloads for consolidation back onto the IBM i platform. (There is no special low-cost Application Server license for anything smaller than a 50-class machine, which is a Power 550 or Power 750 in the most recent lines.)

If we want IBM to remain interested and engaged with the IBM i platform that it controls, it has to make money. There is a vast installed base of old machines out there, and they need to be upgraded. With the iron being relatively inexpensive, I think it would be wise for IBM to do something radically different. I want IBM to stop charging for software based on tiers or users or a mix of the two and scaling it up as the machine gets bigger. This is limiting growth in revenues and has diminished the customer base.

What I want IBM to do is to charge customers by the percent of their database, application, and infrastructure workloads they run on the Power Systems-IBM i combination. The higher percentage of the workloads running on the Power Systems-IBM i combination for each of these tiers, the lower the unit price. Also, to keep things simple, and to encourage consolidation, for this consolidation pricing, IBM would charge a straight flat fee per CPW, regardless of the size of the machine and would similarly offer upgrade paths from any machine to any other machine in the current Power7 lineup.

This consolidation pricing would require for three distinct pricing bands inside of logical partitions: a database SKU, an application server SKU, and an infrastructure SKU. If you run all of your databases on the IBM i box, you get the best price. If you have DB2 for i supporting 25 percent of your database workloads and SQL Server supporting, you are going to pay a higher price for IBM i than the customer who is all-in. And the tricky bit is despite all this, the DB2 for i SKU has to be cheaper than Windows Server plus SQL Server for the same workload so customers are encouraged to move workloads to IBM i. The application server has to be very inexpensive, and available on all machines. And you have to define what an infrastructure server is–Web serving, file serving, print serving, and so on–and then charge a very modest fee for this work.

To really replace Windows at IBM i shops, IBM will have to do one more important thing: Clone the Active Directory server that most shops use to control access files and programs running on Windows networks. Centrify knows a thing or two about extending Active Directory to Unix, Mac, Linux, and other platforms. If IBM buys it, they could probably clone it. Mix in the Xamian Mono .NET runtime and I think IBM could make a run at Windows in its own shops.

Everything I just said about consolidating workloads onto Power Systems machines at IBM i shops works equally well for AIX shops. Which is how we sell this idea to the Power Systems division.