IBM Repositions the Mainframe as Central to SOA

Published: May 9, 2006

by Timothy Prickett Morgan

For the past four decades, the glass house has been called a data center, and the key word there has been center, not data. Well, at least for the people selling systems and servers. The fact that the mainframe continues to drive many billions of dollars in sales a year despite all of the many wrenching changes in hardware and software technology in that time is a testament to the technical and marketing genius of IBM.

Of course, it takes two sides of the bargaining table to make a transaction, and the mainframe has persisted for so long because mainframe shops have a strong desire to leverage the technologies that they know best to do new things. They wanted to do client/server computing or Internet computing a decade ago, and they did that with mainframes. Today, they want to implement a service oriented architecture (SOA) for their application and middleware software, but they also want to have the mainframe be central to that SOA. Plus ça change plus c'est la même chose.

Making the mainframe relevant to SOA, of course, means more than allowing applications written on other platforms using tools for those platforms to access data stored on mainframe databases. This would be sufficient if IBM was only interested in issuing press releases to get mainframe stories published. But what IBM wants to do--and needs to do--is get all of the tools that are being used to create applications based on the SOA approach to run natively on good, old z/OS. Because that way, IBM has a very profitable mainframe business for another five years and everybody at IBM gets paid and some get their bonuses.

And that is what last Thursday's mainframe announcements, made by Jim Stallings, general manager of the System z business, and Steven Mills, general manager of Software Group, here in New York were really all about. The main focus of the event was to make it clear that the mainframe would have all the right tools to allow customers to deploy SOAs natively on the mainframe, whether they want to code in COBOL or PL/I or Java and Visual Basic.

Apparently, a lot of mainframe shops do want to code in COBOL and PL/I, even if they do want to do SOA. Kristof Kloeckner, vice president of development at IBM's Tivoli systems management and security software unit, spends a lot of time in the mainframe world; he cut his teeth on mainframes and was actually in charge of bringing the AIX Unix environment to the mainframe many years ago. SOA may be all the rage now in the press and analyst community, but mainframe shops have already been doing this kind of thing for years--particularly in the financial services market. "I have seen quite a few customers that have developed much of their mainframe applications as services," explains Kloeckner. "Many of these companies have large programming teams, which are still programming in COBOL and PL/I, although some are starting to use Java. Core mainframe workloads like transactional systems are being embedded in customer self-service applications."

Of course, all of the major brokerage houses started doing this when the dot-com boom and day trading took off a decade ago; ditto for the big banks and online banking. But back then, financial services companies generally left their mainframe applications untouched, extracted data from their databases that more or less bypassed the mainframe apps, and put the resulting new front end applications on Unix or Windows servers. That approach is so 1999. Now, IBM wants to bring the Rational development tools, the WebSphere middleware stack, and the Tivoli management tools to the mainframe so much of the spaghetti that goes into making up SOA-enabled applications ends up in the mainframe pot. IBM is projecting that if it gets its SOA act together, the number of transactions running worldwide on mainframes can double by 2010. Hold the price of MIPS relatively constant because IBM has no competition. You do the math.

The Rational COBOL Generation tools were one of the neat things that IBM announced last week. These tools were created in IBM's software lab in Raleigh, North Carolina, and they use a platform-independent programming language called Enterprise Generation Language. EGL apparently has the look and feel of a modern Java and Visual Basic, and it is really a stripped down version of COBOL; if you think "Visual COBOL," this probably conveys the right idea. Anyway, programmers code an SOA application in this EGL, and then run it through the Rational tools, which then spit out a real, full-blown, mainframe-native COBOL application.

EGL is one way that IBM is trying to address the skills shortage in the COBOL arena as hard-core coders are starting to retire. As part of IBM's announcements last Thursday, it reminded everyone that it has initiatives in place to add 20,000 skilled mainframers to the pool and has is working with universities around the globe to get new mainframe coursework available to help train newbies.

Keeping the mainframe relevant for SOA means getting software other than the Rational integrated development environment and related tools running on the mainframe. While IBM's WebSphere Application Server has been native on the platform for years, it is only since the advent of the zSeries Application Assist Processor (zAAP) two years ago that WebSphere was an affordable option. With the zAAPs, IBM charges about a third the price of a regular mainframe engine to run WebSphere and the Java application code that rides on top of it. Kloeckner says that since the zAAPs came to market, IBM has seen a "much stronger adoption" of WebSphere on the mainframe. Which Steve Mills over at Software Group loves to see. Linux and Java now comprise more than 60 percent of mainframe hardware sales. (There's a reason IBM just announced the DB2-assist engines, which are known as the System z Integrated Information Processor, or zIIPs. This is the only way to keep the mainframe growing.)

To help push zAAP sales and get more WebSphere code running on mainframes rather than outboard servers, IBM will in June announce that its WebSphere Process Server for z (which does not hand out lawsuits, but rather allows SOA applications to link back into mainframes) and WebSphere Enterprise Service Bus (which is a funky kind of software standard for plugging SOA code together) will run natively on z/OS. IBM is also readying a z/OS version of its WebSphere Portal Server and will put out a z/OS version of its XML-enabled "Viper" edition of the DB2 database. Tivoli Federated Identity manager has already been updated to link the mainframe's RACF security software to the Tivoli equivalents on distributed Unix, Linux, and Windows systems, but IBM is emphasizing that you can put the mainframe at the center of application and resource access and make use of the native encryption the mainframe has woven into it.

IBM will also be banging on the old "total cost of ownership" drum to make its case for putting the mainframe at the heart of SOA applications. Everybody knows that mainframes are not cheap, but then again, neither are programmers and system administrators. Kloeckner cites statistics that show a mainframe fully burdened with hardware (servers and storage), software (operating system, middleware, and database), maintenance for five years, and administrator costs for five years costs about $4,500 per end user to own and operate; a distributed platform like a Windows or Unix server costs around $8,000 per seat, according to IBM. "A mainframe-based system can support more people and can expand to handle more workloads with fewer staff," says Kloeckner.