Parallel Sysplex - tales from the mainframe

This April marked the fifth birthday of Parallel Sysplex, IBM's maligned, derided and misunderstood System/390 mainframe clustering technology.

The lumbering five-year-old was almost stillborn, and has had a controversial and sickly childhood. Indeed, the environment - which is expensive to support and short of customers - has looked at times like a walking corpse.

Ironically, this recent view of Sysplex was compounded by the success of IBM's new-generation G5 mainframe. Now the G5 is snapping at the heels of more expensive but more powerful Emitter Coupled Logic (ECL) machines, would anybody want to cluster machines which already provide the benefits of a clustered environment?

With no obvious market, and not a lot of existing customers, the question on a few lips is: when will Sysplex be locked in the crypt with all the other zombies? In this penumbral view of the area, IBM never really kill anything off, it just lets them fade away. Look no further than OS/2 to confirm Big Blue's ability to keep products hovering between life and death.

This view is due in no small part to IBM itself. But there's also a lack of understanding of mainframe technology among some analysts, consultants and other 'neutral' observers.

IBM opened the book of Sysplex howlers at the very beginning, inventing the totally unnecessary name rather than adopting more accessible terminology.

Parallel Sysplex (the Sysplex is a portmanteau word standing for 'System complex') is to IBM and its customers what everybody else in the industry calls clustering.

The reasons for confusion multiplied, lending the whole area a touch of graveyard-like mystery. The Handbook of IBM Terminology published by Xephon, a UK consultancy that specialises in IBM and its plug compatible rivals, defines Parallel Sysplex as 'architecture cum bundle of products announced April '94 which allow parallel transaction processing on an ES/9000 (the S/390 predecessor) sysplex. Initially supports up to 32 MVS systems, presenting them as a single systems image.' Snappy it isn't.

Some observers argue that Parallel Sysplex has made virtually no impact on the IBM mainframe community in terms of usage. A Xephon survey last year of mainframe users worldwide, Getting the Best out of OS/390, asked users about Parallel Sysplex.

Only one in 11 users were actually running Parallel Sysplex said the survey, which polled senior IT professionals at 137 installations worldwide.

Half paid for their software through Parallel Sysplex Licence Charge (PSLC) which they saw as a cheaper option to standard licence agreements, without even intending to run it.

Such a low take-up was bound to lead to speculation that IBM would abandon Parallel Sysplex at some point, or at least lower its profile in the crowded IBM range. There are a number of reasons why this is unlikely.

First, IBM needs the technology to be able to compete with the more powerful water-cooled mainframes manufactured by Hitachi Data Systems (HDS). Second, with the battle over software licensing engulfing the industry, IBM is anxious to show that it can offer as flexible a licensing policy as anybody else.

The third reason is less defensive and more in tune with predictions of where the market will be heading in the next few years.

IBM expects a major growth in electronic commerce applications, which will require massive processing power.

According to Doug Neilson, an IBM systems consultant for high-end systems, the reason that IBM has been relatively quiet about Parallel Sysplex is that the technology has matured from the roll-out phase to the business phase.

Neilson says there are now some 80 Parallel Sysplex sites in the UK, of which half are using the system for data sharing - its intended use - rather than just the minimum features required to obtain a benefit from lower licence charges. The worldwide installations number around 1,400, some 40% of which are data sharing, according to Neilson.

Parallel Sysplex was born 'in the ECL era and its implications are at least as significant from the cost point of view as from the technological one, especially for those lower down the scale', says Xephon.

IBM's decision to adopt the Cmos chips as the engine for its mainframes was widely criticised in the early to mid-1990s. The older ECL chips, which had powered IBM and compatible mainframes in the past, were more powerful but considerably more expensive. Cmos, while cheaper, could get nowhere near the power of the ECL-based machines until comparatively recently.

The plug-compatible mainframe manufacturers (PCMs), Amdahl and particularly HDS, had a field day following Big Blue's decision to take the Cmos route.

For that very small group of mainframe users requiring massive power and price/performance IBM's Cmos just could not meet the case.

For several years HDS stayed king of the castle as far as the high-end mainframes were concerned. Parallel Sysplex was IBM's attempt to provide a similar level of power for these high-end users. But only in the last 12 months, with the latest generation of IBM G5 Cmos-based systems, has the company had the power to challenge HDS.

IBM was at first desperate to find a way to match the power of the ECL machines from HDS and Amdahl.

Sav Mellor, a senior product manager with Amdahl, says: 'IBM had a real incentive for going for Parallel Sysplex. Six Cmos engines in total only came to the power of one ECL machine, and IBM had to find a way of offering more power at a reasonable price.'

HDS and Amdahl, with their ECL or hybrid ECL/Cmos offerings, had a four-year window in which to make money by giving IBM users the power they required and charging a premium for the privilege. That window is fast closing. 'Cmos has come a long way from those early Sysplex days and is now more powerful and reliable,' says Xephon senior consultant Mark Lillycrop.

'IBM needs Parallel Sysplex as a way of differentiating itself from the other mainframe platforms. It is the high availability that is important. No one can guarantee 100% up-time, but it is as much a matter of confidence in the system as anything else. Parallel Sysplex offers flexibility and scalability.'

Robin Bloor, chairman and chief executive of consultant Bloor Research, believes IBM's biggest customers are gradually moving toward Parallel Sysplex. 'Interestingly, a number of IBM's top-end customers have moved to Parallel Sysplex for their web sites,' he says.

'I believe that Wal-Mart in the US is one of them. The reasons are fairly simple to understand. First, it offers you 99.999% availability, which you need for ecommerce. Second, the total cost of ownership is cheaper than high-end Unix,' he adds.

IBM has, as Lillycrop points out, gradually improved the price/performance of Parallel Sysplex and Cmos chips to such an extent that the rival mainframe suppliers have almost lost their edge.

Mellor admits that Trinion, Amdahl's latest offering based on hybrid ECL and Cmos technology, has only a year's grace before IBM releases a Cmos machine that matches its power and performance. Bloor concurs with this, as do some of the plug-compatible mainframe suppliers.

'IBM is delivering a 50% improvement in performance every year. IBM made well over $1 billion out of System/390 last year and most of it came from Parallel Sysplex,' says Mellor.

Another reason why IBM needs Parallel Sysplex is that it offers yet a weapon in the battle between suppliers, and those between suppliers and users over software licensing.

The days when IBM and the other mainframe companies could charge for their software based on the size of the processor on which it ran are coming to an end. Under this regime, the bigger the machine the more expensive the software licence, even if the user only used a fraction of the power of the system. But the customer base is no longer willing to pay a premium based on a notional price/performance argument, as Oracle has recently discovered.

The UK Oracle User Group has demanded a fairer and more rational method of software pricing. Oracle, although it will not openly admit it, is being forced to consider such a move.

In doing so it is trailing IBM. IBM's mainframe customers have received a host of software licence offerings for some years. The Parallel Sysplex Licence Charge for example, means that any customer who contracts to take the Parallel Sysplex coupling facility - the key component of the system which allows them at some stage to run Parallel Sysplex - is entitled to special licensing terms. By agreeing to use Parallel Sysplex at some point, you are entitled to the reduction in software charges that PSLC brings.

Following Xephon's revelations that PSCL buyers are not even using Parallel Sysplex, IBM began to insist that customers paying their licences under PSLC actually use the technology.

Lillycrop doubts if this move was seriously aimed at the customer base. 'IBM tightened up a little bit last year, but only to stop the systems integrators taking advantage of PSLC. I don't think that any mainframe user who wanted PSLC would be refused,' he says.

Neilson claims that the number of customers running what became known as a 'shamplex,' has, in fact, been reduced to nothing. Although not all sites are using Parallel Sysplex to its full capacity, and less than half are carrying out full data sharing, many sites are taking advantage of features such as running their systems as a single operational image.

Even IBM will admit privately that it was being a little 'economical with the truth' when it was using PSLC as a basis for assessing the numbers of companies employing the technology. There are signs that the numbers of customers using the technology as intended is increasing, according to Lillycrop and Mellor. 'There seems to be a steady trickle of people moving towards Parallel Sysplex but it is never going to be a huge market,' Lillycrop argues.

Far from abandoning Parallel Sysplex because of the slow take-up, IBM customers are moving closer to the technology Mellor believes. 'We held a customer requirements forum at the end of last year, and spent a lot of time on clustered systems. The message from customers is that you can forget your 30,000 Mips system - what we want is ten 3,000 Mips systems,' he says.

Mellor believes that one of the reasons for the slow take-up of Parallel Sysplex, other than as a way of reducing licence charges, lies in the preoccupations users have had with the year 2000 and the euro.

'We think there will be greater usage in the second quarter of 2000. The reason we think that the growth will be there is that the customers have spent the last two years doing tactical things such as year 2000. Once the year 2000 problems are out of the way then they will return to thinking more strategically.'

One of the reasons that Mellor feels Parallel Sysplex will succeed is the growing need for high-availability systems, with more and more companies involved in heavy-duty transaction processing.

In the past it was easier to take a machine offline for a few hours to install an upgrade but today, with email, ecommerce and other online and Internet applications playing an increasingly important role in commercial life, this is just not possible. High-availability systems have been available since the 1970s from companies such as Tandem. But these were in needed for those applications which had to be continually online, such as banks' automated telling machines.

Indeed, apart from the financial and defence sector, Tandem and its rival Stratus sold into virtually no other markets. Parallel Sysplex now offers the sort of fault-tolerance that only Tandem and Stratus were able to offer in the past. IBM tacitly acknowledged that it had no offering in this arena when, in the 1980s, it decided to re-badge machines from Stratus and sold them as the IBM System 88.

'It is primarily about availability,' recalls Neilson. 'I was taking a seminar with a group of NT consultants and I had to stop my presentation to explain to them how we can replace an operating system without taking the machine down.'

Mellor believes that Parallel Sysplex has every chance of success because IBM has sanctified the technology.

In many ways he sees it as being as important as the 1974 announcement of Systems Network Architecture (SNA), which set the standard for mainframe networking for 20 years - until it was displaced by TCP/IP.

Of course, things have changed since then. IBM is no longer able to dominate the industry as it did 25 years ago, and there have been other IBM architectures since, most notably Systems Application Architecture (SAA), which have failed conspicuously.

'The difference that Parallel Sysplex has made is that it introduced a formal architecture to clustering in the IBM environment. Previously you had been able to cluster systems together but there were incompatibilities,' says Mellor.

IBM is banking a great deal on ecommerce taking off and thereby increasing demand for high-availability systems with the mature and stable environment it views Parallel Sysplex as providing.

According to Bloor, ecommerce is growing rapidly with the number of people buying on the web increasing by as much as eight per cent a month.

The number of companies with profitable ecommerce sites has risen from 33% last year to 46%. Last Christmas web-based sales topped a record $12 billion.

But if ecommerce is to succeed, it requires a high-availability system with full data security and data integrity features. This is where IBM is pinning its hopes for Parallel Sysplex.

'Parallel Sysplex has provided the first real infrastructure for continuous operations in the System/390 environment. If I am going to buy a new car and the web site is down, I am not going to bother to wait to see if it will come up again, I am going to go elsewhere,' says Mellor.

Other systems have been launched in the past which have sought to take advantage of an emerging market. Nobody can claim, least of all IBM despite its protestations to the contrary, that Parallel Sysplex has not looked a bit like a member of the undead at times.

At best, the market is still small in terms of the number of customers who would be interested or capable of running a Parallel Sysplex site.

But the window of opportunity which IBM's rivals had at the high-end is rapidly closing as Cmos power increases.

It may not be the happiest fifth birthday a technology has ever had, but it is by no means as gloomy as those once passed by the architecture's elder brother, SAA, which just got worse year after year. There's more life in Parallel Sysplex than may have appeared at first.