Technology News Technology News

Another Perspective


Far, far from the big iron world of 10,000 MIPS shops there are some very different customers who need just a little mainframe power.  If they get new iron, it will be a z9 BC.  But if they can't afford a new machine or can't pay for the transition to 64-bit software that's required to qualify for IBM's best deals, they have to do something a bit different.  Basically, they have two choices.  They can find a mainframe hosting company.  Or they can ride the wayback machine all the way to a Multiprise 2000.

These small shops formerly had a third choice, a Flex-ES emulated mainframe from Fundamental Software, and if they made that choice in the past they can stick with it.  But if they aren't on a Flex box already, they can no longer move in that direction.  Depending on their current position and their plans, they will have to settle for one of the options still available.

This group of users is not uniform.  Some of them are shops that decided to move off the mainframe to another platform but still need a mainframe for some jobs that haven't yet been transferred over.  Others have successfully moved off the mainframe for production, but they still need a mainframe system in reserve to fulfill compliance requirements that their new systems are unable to satisfy, at least not a cost that makes sense.  Still others are very happy with their small mainframes but own old iron that is no longer easy to maintain, even with the aid of third parties.  And then there are the shops that moved to machines with lots of headroom and realize they will never need the extra power they have, nor the extra software license fees that spare power requires.

IBM has tried to give all these users a way to move to current mainframe technology.  It offers technology to bring the speed of its z9 systems down to very modest levels and priced the governed equipment (and its software) accordingly.  It has offered users who move to new machines caps on their software fees, too, so headroom is not a budgetary headache.  It has made it a lot easier for customers to tuck Linux servers into their big frames, encouraging physical consolidation of server farms.

Nonetheless, the structure of IBM's offers can't always tempt shops running 31-bit equipment that want to leave their software pretty much alone.

To take full advantage of IBM's deepest discounts and other pricing incentives, a customer moving to a z9 has to migrate to a stack of current systems software and middleware.  That migration is a barrier to some of the smaller shops because it requires time, money, and skills that are in short supply.  And, notwithstanding the theoretical benefits of current technology compared to older mainframe environments, a lot of these smaller shops would not obtain any operational benefits from software and hardware upgrades other than the simple advantage of having new equipment versus old.

So, when users of small mainframes start worrying that their old machines have to be replaced, they sometimes look at options that are considerably different from anything their IBM reps will offer.  One choice, particularly for users who require under the 60 MIPS of the smallest Multiprise 3000, is an old Multiprise 2000 or, as IBM calls it, a Model 2003.  (Multiprise 2000 processors are available with speeds of up to 160 MIPS, but few if any users who need at least 60 MIPS would find the machines attractive.)

On the plus side, a Multiprise 2000 is dirt cheap.  There are no firm prices for these rarely traded processors, but it should not be hard for a customer to buy a skinny working machine for a few grand and to get one that has lots of features, including internal disks for a price in the teens.  IBM might be willing to offer factory maintenance on a special bid basis, but anyone who installs a mainframe built in the late 1990s ought to investigate third-party maintenance.  Alternatively, a buyer with sufficient floor space might be able to buy a pair of 2003s and keep one around as a disaster recovery machine (and a source of parts).

If a Multiprise 2000 will IPL, chances are it will run flawlessly for years.  These computers are IBM classics, built to last far longer than their turn at the forefront of the IBM mainframe product line.

One caution for prospective buyers: Some IBM software that is more recent than the Multiprise 2000 requires processor features that were introduced on successor machines.  The Multiprise 2000 is, architecturally speaking, a 9672 G4.  IBM may also decide that it will not permit a customer to move a software stack to a Multiprise 2000.  On the surface, IBM says its will no longer license old software on a fresh (if not new) platform.  But a prospective user isn't, strictly speaking, looking for new licenses, only the transfer of existing valid licenses from one system to another.  What it comes down to is getting approval from IBM.  IBM doesn't want to lose a footprint, but neither is it obliged to support a user living in the past.

Another issue users have to check out is the availability of spare disk drives for either growth or replacement.  The internal disks used in IBM mainframes that provided in-frame storage are not industry standard products.  The internal drives used in Model 2003 systems are unique.  One defensive tactic buyers may adopt is to get extra disks in the first place so their spares are sitting right in the frame.

While there is no question that a user's interest in a Multiprise 2000 is exceptional, the circumstances that make this possibility an option affect quite a few mainframe shops.

One factor is the high price of mainframe hardware compared to alternative equipment with comparable processing power.  IBM mainframes are built to exceedingly high standards.  It is possible that the users of small mainframes simply don't need the kind of belt-and-suspenders technology that their larger cousins consider vital.  All servers, from the most widely used X86 and X64 boxes up to the top-end IBM z9s are pretty reliable, but clearly some machines are sturdier than others.  Every user has to decide how to balance cost and reliability.  But in most cases, shops that need mainframes with speeds under 60 MIPS can find sufficient raw processing power and ample I/O bandwidth in very modest alternative servers.

Another issue for mainframe users at the low end of the customer base is software pricing.  Compared to the systems software on alternate platforms, mainframe software can be expensive.  Users offset this cost (and the price of their hardware) by getting much higher utilization out of mainframes than alternatives.  The net effect is that mainframes can yield efficiencies that in theory could be obtained (and exceeded) by other platforms but which in practice are usually not.

For big mainframe shops, keeping the iron running at capacity is a continual process, and the cost of the systems specialists who do the tuning is budgeted accordingly.  For small shops, efficiency may be the result of past tuning efforts, but budgets for staffers or consultants are tight.  There's a big advantage to keeping a stack of software as undisturbed as possible.  So it is hardly surprising that a small mainframe shop with mature applications would want to stick with hardware that lets it preserve every line of code, every setting, and every other component of production workloads.  Small users worry about the possibility that migration from a well-tuned 31-bit environment into the unknown world of 64-bit technology will lead to unexpected costs.  System efficiency might decline, and that would call for a costly effort to max out the new platform.

Users concerned about the performance of new equipment can arrange for benchmarks.  But, chances are, they can't just load their old applications into a current stack of systems software and middleware and see how it runs.  Migration is a bit more difficult than that.  Even with a lot of testing, predictions involve a degree of risk.  But IBM is unrelenting in its insistence that users who want the best deals on a z9 have to move to current software technology.  And users who think they can cover their bets on a 31-to-64-bit migration by taking advantage of IBM's powerful partitioning technology, usage-based software offerings, and other incentives would be wise to consider the one element of this kind of migration IBM isn't keen to talk about: The cost of bringing all this potential to bear on what is a very small workload by mainframe standards.

So, it's not just nostalgia that gets low-end mainframe users interested in a Multiprise 2000 or hosting on a 31-bit system.  It's the fear of unexpected expenses in settings where the boss might well ask whether his company really needs a mainframe at all.

— Hesh Wiener June 2007

Copyright © 1990-2019 Technology News of America Co Inc.  All rights reserved.