Rivalry and confusion at IBM

Rivalry and confusion between product brands is as strong as ever at IBM. Sean Hallahan separates the strands.

There is a probably apocryphal story that, during the 1970s, an IBM mainframe salesman would nip smartly into a telephone box and reveal the details of the account to a rival at Amdahl. Why? Because he feared losing the account to his supposed colleagues in the IBM mid-range division.

Internal competition at IBM was so intense that the mainframe division often felt less animosity towards its opponents than towards colleagues.

But in those days, the division between mainframe and minicomputer meant something.

The mainframe was located in a data centre and handled massive workloads for applications such as on-line transaction processing for an ATM network. The minicomputer, by contrast, was designed to function as a departmental or medium-sized business machine.

The appearance of the PC in the early 1980s and its subsequent growth in power and functionality sowed the seeds of the confusion that surrounds IBM's hardware strategy today. The problem is that there is no clear delineation, either by application, hardware or operating system capabilities, between the mainframe and the minicomputer, and between the low-end minicomputers and the top end PCs.

Nor is price any longer a reliable indicator of the differences between technologies. The lowest class of AS/400, costing about #5,000, is no more expensive than a high-powered desktop machine. The same applies at the top of the AS/400 scale, where the price and power of a so-called mid-range system exceeds that of a low-end System/390 mainframe.

This blurring of the differences between the various systems has led some analysts to speculate that the IBM mainframe as we know it today, based on CMos chips like the ones in AS/400 and desktop systems, will no longer be the basis for the mainframes of the future. IBM's large systems for the next century, so the thinking goes, will be based on AS/400 architecture.

One analyst who believes this is Phil Payne, managing director of Isham Research, but he also detects signs of a fight back in the System/390 camp. 'In general I still hold the view that the large systems will be built around the AS/400 but what is interesting is that the launch of G5 (the fifth generation of IBM CMos chips) last year took everybody by surprise because of its power - including IBM itself. Outside the semiconductor labs, no one knew what was coming,' says Payne.

The power of the G5 is a contributory factor in galvanising the IBM System/390 division into action, Payne says.

'There are signs that some teams at IBM are showing they can be more responsive to customers' needs,' he says.

s an example, he cites customer reaction to automated software fixes which IBM supplies to its users. 'Some of the users have said they would still prefer a core dump (code download) because they can track the code and sort out the problems themselves, and IBM has told them how to disconnect the feature,' Payne says.

But despite IBM's new-found attentiveness to customers' wishes, Payne still believes the AS/400 is the technology that is most likely to lie at the heart of the next generation of large systems.

ccording to Payne, there are good reasons why the AS/400 is better than System/390. It is already a 64-bit technology; it is highly scalable; it has Java embedded in microcode, making it a powerful Internet machine; and it has no plug-compatible competition either for its processor or, perhaps more importantly, for the AS/400 storage devices.

There is also no real competition in the database market: the AS/400 operating system has its own version of DB2 embedded in it, vastly improving database application performance. You can understand your bills, too, which is always welcome. There are plenty of developers who charge for their software on a per-seat basis, like IBM itself, rather than in the chaotic way in which the mainframe software companies operate. One of IBM's major problems with developers in the System/390 environment is software licensing.

To be fair, IBM has tried to make its licensing policy more flexible.

In the past, the size of a software licence was dictated by the size of the processor: the bigger the machine, the higher the licence fee. Unfortunately the process was discriminatory in the extreme. If a customer purchased a larger machine to cope with abnormal workloads at certain times, but ran the machine at only 75% of its capacity for most of the time, it was still charged on the assumption it was running the machine to the red line.

IBM tried a variety of schemes to redress the balance in favour of the customer by charging only for the parts of the system actually in use, but the message did not seep through to many of the application developers, who continued to charge for their products on the old capacity-based licensing method.

Since it is widely estimated that software charges amount to at least 50% of data centre costs in a mainframe environment, these charging practices inflate system cost considerably. This is also widely reckoned to turn many a potential System/390 customer toward the AS/400 or another hardware base.

Of course, IBM is not alone in having problems in the mainframe environment.

ICL, Bull, Unisys and other vendors of large systems have their problems, too, but IBM's are worse than most. First, it has the problem of having to fight off pressure from IBM-compatible suppliers such as Hitachi Data Systems (HDS), Amdahl and the storage supplier gang, led by EMC. Second, it has what the likes of ICL, Bull and Unisys do not: internal competition, particularly from the AS/400 and the RS/6000 Unix range.

ICL recognises that selling mainframe systems is neither as easy nor as profitable as it once was. Peter Slavid, ICL's business strategy manager, declines to comment directly on IBM's fortunes, but offers some indication of how ICL sees the market.

'As far as we're concerned, the bit of the mainframe that is important is the operating system. Mainframe hardware and operating systems were designed together. We have taken a decision to shift VME (ICL's mainframe operating system) over to the Intel platform. That enables us to retain the customer- built applications running with transaction processing monitors and probably hierarchical databases and run them on Intel-based systems.

The hardware does not really matter any more,' Slavid says.

Slavid is well aware of many customers' reaction to the relatively high cost of mainframe systems. 'Some customers say to us, "we know this solution is better and that solution is inferior, but it is still too expensive", and there comes a point when it's not worth pushing. You can see the same sort of issues being discussed at IBM,' he says.

IBM claims to have sold more mainframe processing power last year than at any time in its history, and there is no reason to doubt this statement.

'We shipped twice the number of Mips in 1998 as we did in 1997, and we will not stand still on our need to be attractively priced,' says Dave Carlucci, IBM general manager for System/390.

The difference is that IBM is not reaping the price-per-Mips that it would have had a decade ago. Today, all the profits are in software and services such as consultancy, maintenance, outsourcing and systems integration.

IBM is not the only company to feel the pinch as margins on tin fall.

s a result, what used to be a service has become another charge on the customer.

'Hardware margins used to be very high, so we could afford to have an engineer on site. If the customer had a problem we would send in a consultant free of charge,' says Slavid. But although hardware maintenance costs are now chargeable, air-cooled Cmos-based systems are more reliable than the water cooled ECL (emitter coupled logic) machines. That means that revenues accruing from maintenance are low, according to Payne.

Carlucci reckons one of the key factors in the growth of the System/390 business was IBM's decision to consolidate its server product sales.

Instead of having four different IBM salespeople - one from System/390, one from AS/400, one from RS/6000, and one from Netfinity - trying to sell a server, a single salesman will offer all four products. But even IBM concedes that there is a degree of confusion over its policy on large systems.

'It is clearly not that clear, although that is a clumsy way of putting it,' admits Peter Norris, IBM System/390 consultant for Europe. 'Netfinity is aimed at the NT market, which is growing bigger all the time. There are two machines in the middle: the AS/400, which is a machine people love and which has an immensely strong following; and the RS/6000, for people who want Unix, and good old-fashioned System/390.' But Norris admits that the lines between the lower, middle and upper sectors are becoming blurred.

The official IBM line is that it has a product and a price to suit every application from the desktop to the data centre. But that arguably does not make Big Blue unique. Sun, which recently launched a machine it claimed was capable of running at 3,000 Mips (a claim Norris is sceptical of), also has machines that it says operate from the desktop to the data centre.

The difference is that Sun has only one operating system, Solaris, while IBM has perhaps a dozen.

HDS has just launched a 3,000 Mips machine, the Trinium, which makes it the most powerful beast in the IBM-and-compatible stable.

Unlike IBM, HDS still builds systems which combine Cmos and ECL technology and are more powerful than anything that IBM can offer. But Dick Poynter, HDS field product manager, acknowledges that the gap between the two technologies is narrowing.

'The System 390 is doing very well but the AS/400 is increasing in power all the time, and is certainly something serious users would consider purchasing. The Trinium is the most powerful machine on the market, but given the fact that Cmos performance is increasing by 30% a year, it probably gives us only a three-year lead,' Poynter says.

IBM, along with HDS and other large systems suppliers, is looking to the next wave of business change to ensure the survival of its systems.

Ecommerce, data warehousing and other power-hungry computing tasks are creating growing markets. But this may not be enough to ensure the survival of the System/390 in its present form, says Payne. 'A lot of data warehousing number crunching is being pulled off the System/390 and carried out on an RS/6000 SP2,' he says.

But there certainly has been a return to the mainframe and to the more centralised computing model that a minority of analysts have been predicting for some time. Most hardware vendors have only one offering in the mainframe class. But at IBM the RS/6000, the AS/400 and the System/390 all offer systems powerful enough to be classed as enterprise systems.

Despite the fact that the same salesforce is marketing all three lines, the underlying rivalry which has been part of IBM's culture since it had four different divisions is as lively as ever.

Buying from IBM, or planning an IT strategy as an IBM enterprise systems customer, is still as difficult and confusing as ever.

THE IBM SERVER RANGE

In a world of interoperability, platform-independent software, and bandwidth everywhere, IBM's once very different children are beginning to look far more similar in terms of capability.

Netfinity is an Intel-based server, nicely engineered, not radically dissimilar to other Intel-based servers from Hewlett-Packard, Compaq, et al. Just more black.

S/400 is IBM's very reliable proprietary mid-range server. AS/400 has always had its own version of DB2 embedded in it, resulting in excellent database application performance. Transaction processing software, other middleware, and now Java are also embedded at the most fundamental level.

mazingly, it was widely tipped for discontinuation as recently as 1994.

RS/6000 is Big Blue's Unix server and workstation range, covering desktop boxes and powerful parallel SP servers running IBM's proprietary flavour of Unix, AIX. An RS/6000 called Deep Blue famously beat world champion Garry Kasparov at chess in 1996.

S/390 is the IBM mainframe which has seen a return to popularity recently.

It looks like the AS/400 architecture may surpass its abilities in five years or so however. S/390 is not so easy to upgrade as rivals from Hitachi and Amdahl.