IBM is revamping its entire server line in a repositioning exercise that includes everything from Windows NT machines to mainframes, and confirming that the AS/400 is now IBM's long-term platform of choice.

IBM started merging the technology of its S/390, AS/400 and RS/6000 machines last year and the AS/400 and RS/6000 already share chips and electronics.

But the AS/400 remains IBM's mainstream high end platform, and the first fruits of this renewed focus will appear in a couple of weeks, when IBM will introduce an AS/400 baby mainframe with logical partitioning, Netfinity links and an Intel Xeon PC card.

Although IBM expects S/390 to be around for a long time yet, it accepts that if a company has not implemented one by now, it is unlikely to in future. As a result, IBM recognises that to win new mainframe business today, it has to penetrate its existing mainframe base with additional systems, applications and services.

Phil Payne, analyst at Isham Research, also believes that IBM sees Windows NT connectivity to the S/390 as crucial to future growth. The problem with NT is its scalability, he noted, which is where S/390 and the AS/400 come in.

"IBM won't criticise Microsoft publicly, but privately, it acknowledges that serious technical issues limit NT's scalability," he said. "NT 2000 might scale up to five or six processors, but corporate users need to go up to 64 or 128, and that means clustering technologies."

He continued: "IBM's idea is to attach a Netfinity bus to the AS/400 and use the AS/400 as a management platform for NT. It's a huge market opportunity."

Where does the S/390 fit into IBM's much-hyped "Ebusiness" strategy?

It undoubtedly has an image problem. It is seen as expensive, and while the platform currently remains at the heart of corporate computing, it may, in future, be overshadowed somewhat in performance terms by Hitachi's next generation Skyline system. This is expected to be the world's most powerful mainframe.

The first Skyline generation bit deeply into IBM's mainframe business as Big Blue made the painful transition to industrial strength Cmos and, until recently, that mattered enormously. The mainframe market is power-hungry, and whichever supplier occupied top-gun position had a huge marketing advantage.

Today, many of the old drivers have fallen by the wayside.

IBM announced its Sysplex mainframe clustering technology in 1994, but the rest of the world and Big Blue's ISVs needed time to catch up. Four years later, such software inertia no longer holds sway, however, and ISV applications are Sysplex ready.

Sysplex is to processors what Raid is to storage. If a central electronic complex (CEC) falls over, it is no big deal. It is no longer a show stopper.

Take, for example, Company X. It needs the equivalent of 10 engines of new G6 capacity in the third quarter. It can buy one 10-engine system from Hitachi Data Systems, or two 5-engine systems from IBM. With Aggregated Sysplex pricing, there is no difference in software costs since the charge is per MSU (main storage unit).

There are other advantages to having more than one mainframe. With two boxes, companies gain more organisational and business flexibility, and they can power down when the business no longer requires the extra power.

Market demand for Skyline had previously been based on a combination of factors: the need for power; the punitive costs of additional resources; and the inertia surrounding Sysplex.

Today, the argument is about which supplier can offer the most effective clustering technology, and the fastest bus and most bandwidth. The organisational requirements for internal and external bandwidth are growing quickly with the spread of Lans, Wans and extranets, and IBM appears to be in a strong position to meet the new demand.

There are still some compelling reasons to purchase a Skyline, and Cics IDMS is a classic example. While ISVs have dealt with the inertia surrounding Sysplex, many customers have said they will not migrate to Sysplex-capable software until after 2000.

As board directors demand that IT departments focus on essential Year 2000 compliance issues, organisations are operating on frozen software budgets. They are unwilling to tackle the millennium bug problem and also run the inevitable risks involved in implementing new software, particularly if the application is as business critical as IDMS.

As the Year 2000 problem is resolved, however, Sysplex-capable software will be installed more widely and, as applications reach the end of their natural shelf life, IT departments will have the time and resources to implement new applications.

Also, while the next generation Skyline machines may offer scorching performance, IBM's S/390 scalability plans are equally impressive. By 2001, the machine is expected to scale by a factor of three; the AS/400 and RS/6000 by four; and the Netfinity family also by three.

According to Isham's Payne, IBM has set itself the corporate goal of being number one or two in every TCP benchmark test over the next five years.

He said: "IBM believes that Mips on demand is an essential and unavoidable component of any machine used in a business environment and it recognises a critical issue with electronic commerce is sizing the systems. You just don't know if you'll get 10,000 or a million transactions."

To achieve this, IBM has two choices. One would be to formalise the terms and conditions that would guarantee Mips on demand.

The other approach would be to use Very Long Instruction Words (VLIW) technology, deriving improved Merced performance from the quality of an IBM compiler. Of the latter, Payne noted: "It not only improves performance relatively cheaply, but it differentiates Netfinity from the rest of the Intel pack. The S/390, AS/400 and RS/6000 are clearly differentiated, and this marks out IBM's approach to NT and Netfinity."