IBM's Unix roadmap

IBM's takeover of Sequent last September presented the company with a problem: what to do with two very different Unix product...

IBM's takeover of Sequent last September presented the company with a problem: what to do with two very different Unix product ranges

Sequent came to IBM with a single, recently introduced product line called Numa-Q, which runs the company's proprietary version of Unix, called Dynix/ptx. Numa-Q, a successor to the earlier Symmetry range, was launched in February 1997. It is built from Intel processors, but is distinctively different from most other Intel based servers, because of its Numa (Non-Uniform Memory Access) architecture (see panel).

Sequent's strategy was to appeal to enterprise-level customers, and the company had become the market leader in this sector (for Unix systems worth $100,000 and upwards), after competing effectively with the likes of Hewlett-Packard and Sun, not to mention IBM itself. The success had come from concentrating on certain key applications, such as data warehousing, ERP, and more recently CRM, and from building an effective services organisation to support the product line.

Much of this was in line with IBM's own direction. Big Blue itself is strongest in the large enterprise part of the market, and has increasingly been seeing its competition here as being HP and Sun, rather than its traditional rivals, Amdahl and Hitachi. IBM has also been concentrating on building up its services organisation to be the engine of future growth.

So, as John Pattenden, product marketing manager with formerly Sequent and now IBM, puts it, there was 'cultural synergy' between the two companies. When Sequent made a trading loss in 1998, the share price plummeted, and it became vulnerable to takeover. IBM was an obvious choice, and for once the expected happened, and the two companies did indeed join up.

IBM's first step after the acquisition was to turn Sequent into the IBM Numa-Q Division, a clear statement that Sequent's technology was to be merged completely with IBM's (in contrast, the company's two previous major acquisitions, Lotus and Tivoli, have both retained their existing corporate brand names to this day).

The Numa-Q product line, meanwhile, became a fifth IBM server brand, alongside System 390, AS/400, Netfinity, and IBM's own Unix brand, RS/6000.

Many analysts have argued that four brands are too many: five certainly are, especially when two of them both run variants of Unix. IBM has now taken the first step towards reducing the five, by merging the Numa-Q Division with the RS/6000 operation, to create the Web Server Group. The new name was chosen, according to Pattenden, to emphasise the major application area the Unix systems are designed to address.

As for the future, Sequent already had a roadmap which contained two key elements. The first was to develop Numa-Q, so that it could support Windows NT, as well as Dynix, using the brand name Numacenter. This was introduced in March 1998 and, according to John Pattenden, by the time of the merger 'many' of these systems had been sold.

Pattenden sees this type of development as the future for every vendor. 'Virtually every type of system will have to deploy multiple platforms and multiple operating systems.' Many other companies are already moving down this road, including Bull, Data General, ICL, and Unisys, while Hitachi has recently announced a similar development under the codename Hercules.

The second element was recognition that Sequent could not afford the development effort needed to produce a 64-bit version of Dynix. The company therefore looked around for partners, and after an unsuccessful liaison with Dec (stymied when the latter was taken over by Compaq) decided to team up with IBM and SCO, as a partner in the Monterey project. That was in October 1998.

Monterey is a development effort designed to produce a 64-bit version of Unix that is backwards-compatible with both Aix and UnixWare. Sequent's part has been to contribute expertise derived from its own innovative technology or, as Sequent product marketing manager Dave Burgess put it at the time: 'Our role is to supply IBM with Numa technology, and operating system support for 50,000 users.'

Sequent had already taken two steps towards Monterey before the acquisition by IBM. In January 1999 it announced UnixWare ptx, which started the process of aligning Dynix with both UnixWare and Aix. In the second half of last year it introduced a second release of UnixWare ptx which shares common programming interfaces with UnixWare and Aix.

So some of the groundwork for the convergence of Numa-Q and RS/6000 had already been laid even before the merger took place. Meanwhile, IBM was developing its own Power microprocessor technology for use in RS/6000 (and AS/400), and announced the Power4 evolution shortly after the Sequent acquisition.

The Power4 chip, which will appear in servers in 2001, features two independent processors, each operating at faster than 1GHz, sharing a Level 2 cache. It employs IBM's copper technology for all the interconnects.

IBM's roadmap for the future is essentially to bring all these technological developments together into one box. As John Pattenden says: 'We will offer choice of hardware, choice of operating system, and also choice of system architecture. We want to make sure there is no complexity in what you can do.' And when will we reach that point? 'The ultimate functionality will come in two years time.'

The first step will come this year, the introduction of a Numa-Q system built from Itanium chips - Intel's first 64-bit processor. That will be capable of running 64-bit NT; it will also be capable of running Monterey. The plan is to employ partitioning to allow users to run both at the same time in a centrally managed way.

According to Peter Norris, IBM System 390 Consultant: 'Monterey will give birth to three parallel operating systems, making it easy to port applications between IA-32, IA-64 and Power technology.'

He goes on: 'One of the things IBM is looking to do is to unify these technologies into a single box at some point in the future. So the NUMA roadmap and the RS/6000 roadmap converge at some point down the road into a single physical bit of hardware with three chip technologies.'

So in around 2002, Numa-Q and RS/6000 will merge into a single product. This will employ both Power and Intel processors, and will run Monterey and 32-bit and 64-bit Windows NT, at the same time if required. It may also run other operating systems, such as Linux.

Missing link

Numa (Non-Uniform Memory Architecture) is a way of linking multiple processors together that is distinctively different from the two commonest ways, SMP (symmetric multiprocessing), or MPP (massively parallel processing).

With conventional SMP, the number of processors is limited by the size and speed of the backplane and the shared system bus. The maximum number of processors IBM offers in its System 390 range, for example, is as small as 12. Clustering (as in a parallel sysplex), allows you to increase that number enormously, but requires much greater management capabilities and load balancing features - one reason why parallel sysplex has taken so long to mature.

MPP systems (such as the RS/6000 SP) allow you to connect hundreds of processors, but message-passing between them is cumbersome, and is orders of magnitude slower than SMP. MPP systems thus lend themselves to applications where there are massive amounts of I/O, followed by fairly small amounts of processing (such as video serving), but not to applications that need to scan large databases in an unpredictable fashion (such as decision support), or that require many random updates involving locking (such as OLTP). MPP systems have also proved difficult to program.

The Numa concept, which originated at MIT, is an attempt to get the best of both these worlds, while avoiding the worst of them. Memory is distributed around the processors, rather than being shared between all of them, as in an SMP system.

It is distributed in such a way that access to the physically nearest memory takes less time than access to memory further away (or in technical terms, memory access latency is non-uniform). Access to memory is via a high-speed interconnect, rather than a system bus. Although distributed, the memory is still managed as a single logical object, and at a hardware level for maximum performance. Numa systems are programmed in the same way as SMP systems. Like SMP systems also, they can be clustered.

Sequent is one of three companies that have concentrated on Numa technology: the others are Data General (now part of EMC), and Silicon Graphics.

Read more on Operating systems software