From Computer Finance, a sister publication
Industry analysts have long identified a trend towards software becoming an increasingly high proportion of the total cost of operating a mainframe system. According the Meta Group, for example, by 1998 software will account for more than hardware in the total cost of ownership.
Close analysis
While combined hardware and software bills represent less than half the total cost of ownership of mainframe systems (with operations taking the majority), close analysis of the increasingly complex terms and conditions for both IBM and ISV (independent software vendor) licenses could well yield some substantial cost savings. The introduction of CMOS processors and Parallel Sysplex from both IBM and the plug-compatible manufacturers of Hitachi and Amdahl has led to an increasing level of complexity in software pricing. Now that sites have several different ways of achieving a certain level of mainframe MIPS power (ECL, CMOS, Parallel Sysplex or a combination), numerous software pricing configurations are available. Some are openly offered by IBM, while others need to be sought out by users. A dilemma currently facing many sites is whether to upgrade existing ECL systems or migrate to CMOS. In many cases, software pricing is the key factor in determining which approach is to be favored. The obvious technical choice (move to CMOS) is, in many cases, far more expensive overall, than upgrading an ECL system. The problem occurs particularly with IBM’s top mainframe Model Group 80. Under ECL pricing, sites are permitted to upgrade their machines in several steps without incurring additional software charges. The 211 MIPS Model 9021-941 is in the same software charging band as a 465 MIP 9021-9X2 system, for instance. Under IBM’s new PSLC (Parallel Sysplex License Charge), model group increments are much closer together. Sites upgrading from an 80 MIPS CMOS 9672-R52 to a 161 MIPS 9672-RX3 would go from software Group 40 to Group 70.
An upgrade dilemma?
This creates a difficult mathematical equation for sites requiring more MIPS immediately. On the one hand, they are well aware that CMOS represents the future of the mainframe. By the year 2000, ECL systems will cost so much more to maintain and support compared to CMOS that they will be all but extinct. On the other hand, meeting short-term MIPS requirements with an ECL upgrade can often be the cheaper option. The actual hardware costs of ECL and CMOS are approximately the same (around ú10,000 – ú13,000 per MIPS) but the real difference occurs in software licensing. Take for example, the case outlined in Table 1 of a site with a 286 MIPS Amdahl 6670M mainframe that requires an extra 70 MIPS. The choice is to link the 6670M to a 73 MIPS IBM CMOS Model R42 in a Parallel Sysplex configuration under PSLC pricing, or upgrade the machine to an ECL 367 MIPS model 8670M and use partitioning techniques to manage the extra resources in a single image.The monthly software, maintenance and power/cooling cost of the ECL upgrade option would be approximately ú123,000. However, the CMOS upgrade would cost about ú130,000 – or ú7,000 more. This occurs because even though the ECL upgrade costs an additional ú10,000 per month or so in maintenance, power and cooling, it incurs no further software charge. By contrast, even though the CMOS machine will cost just ú2,000 a month in maintenance, power and cooling, it incurs an additional ú15,000 a month in software charges. In other cases, where both ECL and CMOS upgrades incur software charges – such as a migration from an IBM 9021-822 to a 831, or the addition of a CMOS model R32 to the 822 – the CMOS is the better option on a pure costs basis. Because of lower maintenance, power and cooling charges the site would save around ú5,000 per month. IBM is slowly realizing that the current software pricing system presents too many inconsistencies and often goes against its own strategic plan – to promote CMOS. Over the next 18 months the company is expected to completely revamp its software model group system and replace it with a variety of user, usage and per MIPS-based pricing methods. The final factor in any equation is the residual value. Here again, there is not a clear case for upgrading with CMOS. Although ECL-mainframe residual values are in free-fall, there is strong evidence that the first two generations of CMOS processors (offering approximately 13 and 23 MIPS per engine, respectively) will hold their value little better as they will quickly be considered under-powered compared to the 40 MIPS models that will come online from all three mainframe suppliers during 1996. Analysts expect current 9672-based CMOS systems to loose about 70% of their value between the fourth quarter of 1996 and fourth quarter 1997. The 40 MIPS machines due out in the summer of 1996 should hold their value rather better.
The ISV pricing position
Sites must also take into consideration the pricing of their ISV software when considering a mainframe upgrade. All mainframe software suppliers are currently re-evaluating their pricing structures in response to IBM’s revised pricing schemes. Once again, choice of model in an upgrade can cost or save a site tens of thousands of pounds. A good example of this is Computer Associates. It has recently restructured its software pricing across the board, and once again, this tends to favor upgrading existing ECL CPUs rather than those that are supplemented with a new CMOS system. The company has recently increased its software charges by up to 10% at the high end, but continues to cap prices at IBM Model Group 100 within ECL. If a site increased the aggregate MIPS in use to the equivalent of Model Group 100 by the addition of a CMOS system such as an IBM Model RX3, software costs would rise by around 15% – as much as ú450,000 over five years. Contrary to IBM’s intentions to encourage the use of the Parallel Sysplex technology (even if only one of the machines has the special software and it is not implemented), CA’s pricing substantially penalizes those sites trying to take advantage of IBM’s lower pricing structure. CA has recently increased the one-time charge for adding a CMOS CPU by 40% even under PSLC. All this adds up to it costing more for a site to add a Model Group 40 machine in 1996, than it did a Model Group 60 machine in 1995. Analysts say that CA’s tactic (and other ISVs are expected to follow suit) is to force users into signing simpler – but presumably more lucrative for CA – MIPS-based deals in 1996 and 1997. These types of deals may appear attractive initially, but are likely to be more expensive in the long-run, according to industry analysts. This is not only because of the high expected MIPS growth of the next five years, but also the fact that few MIPS-based deals build in substantial reductions as sites take on more capacity. Mainframe sites are therefore unlikely to have the opportunity, as under traditional pricing structures, to be able to add 20 or 30 more MIPS without it affecting the software pricing group. Many will therefore incur additional costs. The exception to the ‘big is often bad’ rule, in many cases, is IBM itself. The company terms and conditions are substantially more flexible than many ISVs for application and system software. IBM allows users to transfer software, outsource it, return to regular ownership or move it to equivalent machines without incurring the punitive charges often made by other large suppliers. Although sites should not see this as a reason to buy IBM’s sometimes ageing software above a best-of-breed solution, it is an important factor if all other things are equal.
Software pricing in dispute
Overall, analysts are extremely wary of vendors attempt to ‘simplify’ mainframe software pricing structures and therefore encourage organizations to consolidate their applications around a few ‘mega’ vendors. Quite simply, the larger vendors tend to be more expensive, more cynical and less flexible in their attitude to software pricing. According to the Meta Group, 20% of mainframe sites have had a dispute over software pricing with a large ISV worth over $1 million, while 5% have argued over sums in excess of $5 million. The experience of large sites is that there are little or no savings in management costs by reducing the number of ISVs in the mainframe software portfolio. Indeed, Meta Group research suggests that substantial savings can be made by investing in a dedicated software portfolio manager to deal with ISV relations. Meta believes that 70% of organizations employing such a strategy could save as much as $300,000 on the annual software budget – enough in itself to fund the position for at least two years. In only 15% of cases did Meta find that these staff failed to generate enough saving to pay their own way.
