Is the Price Right?

Software Magazine, August, 2000 by Barry J. Graham

Mainframe software users need a new model

THE MAINFRAME SOFTWARE INDUSTRY is less than 30 years old, but it has grown to one of the largest and most profitable industries in the world. Software prices are based on the horsepower (MIPS) of the hardware for which the software is licensed. Use more MIPS, pay more money.

In 30 years, this model, considered simple to enforce, has become overpowering, and current and future projected costs of mainframe software threaten the whole enterprise.

Beating Inflation, or Being Beaten?

Today, IBM constitutes over 60% of the mainframe software market by revenue, and for this reason, most independent software vendors (ISVs) tend to copy IBM's pricing practices, though they are not always very fast to move to new structures when announced--unless it suits them! To understand trends in mainframe software, first check what IBM is doing, and then check the ISVs, companies like CA, SAS, SAP, and Oracle, which are likely to be following IBM--albeit at a distance.

Since 1980, the IBM software cost per MIPS for the average user has remained unchanged at around $2,000 per MIPS, which sounds like inflation has been avoided. But look again. A user with 10 MIPS in 1980 probably uses 2,000 MIPS today. So far from beating inflation, that user's annual IBM software cost has increased from $20,000 per year in 1980 to $4 million today.

This represents an average growth rate of around 30% per year in the installed MIPS over a 20-year period, which is roughly six times the rate of business growth during that time. The same factors are expected to accelerate even more during the next five to 10 years, to at least 50% increased capacity per year. Consequently, by 2005, the 2,000-MIPS user is likely to have 14,000 MIPS installed, and face an annual IBM software bill of nearly $28 million.

Along with a growing IBM software bill, the average user's ISV software bill has risen from $1,000 per year per MIPS to $1,500. Translation: Over the past 20 years, a 10-MIPS user has seen ISV costs increase from $10,000 per year to $3 million, and by 2005 will face an annual ISV bill of close to $21 million.

Masking the Perilous Rise of Software

The real cost per MIPS combines hardware and software costs. While hardware was more expensive, the cost of the software remained relatively flat. But once hardware prices fell below the software cost per MIPS, the overall decline in MIPS price ceased.

The chart, "Change in Costs Over Time" (p. 15), looks at the typical market purchase price for IBM mainframe hardware MIPS and the corresponding total IBM and ISV four-year software costs over the 1990 to 2008 period. For the first few years of this period (and all previous years), the hardware cost was so much greater than the software cost that any reduction in hardware price significantly reduced the total cost per MIPS. But by 1997-98, the software cost overpowered the hardware cost per MIPS. Consequently, any hardware price reductions, no matter how large, had little impact on the total cost per MIPS.

The impact of this can be seen more clearly by looking at the next eight years, when the total cost per MIPS will reduce only marginally from $15,700 per MIPS to around $14,200 in 2004, and then actually begin to increase, due to the ISV cost per MIPS rising gradually over time. (See "Total Four-Year Cost Per Mips," bottom.)

IBM Responds With Programs

In the '90s, suppliers began responding with incentives to keep users on the mainframe. Best known is IBM's Parallel Sysplex License Charging (PSLC), introduced in 1994, where users could combine the capacity of all of their systems on one site, and pay a license fee accordingly. This corrected a long-term abnormality of the mainframe software pricing model, where the software cost per MIPS reduced as the capacity of the system increased, until the user exceeded the capacity of the largest system, at which point the user added a second (and third, and more) system. The cost of each additional system was the same as the first, with just a 10% discount. The cost for two of the largest systems was around 190% of the cost for one. PSLC reduced this cost to 150%, with further savings as the number of combined systems increases. (See "Comparison of Old IBM Software Costs With PSLC Pricing," p. 16.)

Despite the Sysplex model, one glaring problem remained. Users wishing to add a new application to their systems still had to pay the software cost for new capacity for all current products--even if the new application uses few of them. New pricing trends attempted to deal with this.

In one effort, alternative hardware vendors (Hitachi Data Systems and Amdahl) introduced a way of splitting a single system into smaller ones. In effect, these are "virtual" servers with discrete serial numbers that can be verified by the ISVs' software products to ensure that no illegal use is made of their products. To date, the major ISVs have "signed up" for this mode of operation, but as IBM does not (yet) provide the facility on their mainframe processors, it benefits no more than 20% of the user base today. However, by monitoring which programs run on which systems, and how often, even the other 80% can negotiate better software deals for products that don't use the whole system--but only if they use a tool such as SoftAudit from New York City-based Isogon Corp., or other auditing tools, to provide input to the negotiation process.