Running z/OS on x86: Should IBM release a mini-mainframe?

Would running z/OS on x86 with a "mini-mainframe" allow users the benefits of mainframes with the simplicity of distributed systems? In this tip, mainframe expert Robert Crawford makes the case for x86 mainframes and discusses how IBM could approach this project.

Distributed server systems keep getting cheaper. Mainframe software (by that I mean z/OS and related subsystems) continue to be very expensive. If IBM is truly interested in the survival of what it often describes as its "flagship operating system," I think the company should look into moving it onto other hardware platforms.

The "mini-mainframe"
The idea is to create "mini-mainframes," or z/OS running on x86, aimed at a departmental or office level -- areas currently dominated by Windows, Intel and variants of Unix. The idea is not new. Years ago, I saw an IBM demonstration of OS/390 running on a desktop computer. Third parties have created MVS-compatible hardware that ran pretty well until IBM stopped them. Then there's Hercules, an open source emulator, which claims to be able to simulate mainframe operating systems from now and back to the 1980s.

To break into this market, an x86 mainframe would have to meet a few requirements.

First, the mini-mainframe should be able to run on cheap x86 hardware in an air-cooled box small enough to fit under a desk. Packing z/OS onto x86 shouldn't be that hard, although it might require some special chipsets to simulate the mainframe environment. Every computer needs to talk to the outside world, so there should be ports for external disk drives, communication interfaces and, God help us, printers. IBM might also provide a rackmounted or blade version for the modern data center.

Ideally, the z/OS Logical Partition (LPAR) in the x86 mainframe will be a monoplex, greatly reducing serialization stress. However, today's subsystems depend on coupling facilities (CF), which means the x86 mainframe will have to supply one logically or literally.

Second, Web applications will be the primary workload for the mini-mainframe. While 3270 applications aren't automatically precluded, it would not be the interface of choice for smaller shops, and Web User Interface (WUI) applications are the IT industry's direction. This would also put the x86 mainframe squarely in position to run any Java applications and open source already on the market. If a shop insists on using green screens, 3270 access can be easily provided through Telnet 3270.

A good WUI might also avoid the necessity of inventing an MVS desktop for the x86 machine. But at some point, IBM may consider adding a Windows emulator into z/OS just like today's Unix System Services as well as an interface able to interact between the WUI and Windows application.

Finally, the entire x86 mainframe should be sold as a turnkey package, possibly even an appliance. To run with as little fuss as possible, all the software, from operating system to subsystems to applications, would have to be carefully packaged, tested and integrated. In addition to the z/OS, it should include a transaction processor (WebSphere or CICS) and a database management system (DBMS) like DB2, all in a high-availability, fault-tolerant configuration. Some sort of HTTP server would be necessary to support both an administrative WUI as well as the Web applications the enterprise wants to run.

Packaging wouldn't be limited to the systems software. IBM could offer complete packages, providing, for instance, an x86 mainframe with everything an accounting department could need, complete with all the supporting subsystems. Ultimately, the whole bundle could be carefully tuned for the specific application.

Most importantly, the mini-mainframe concept needs bulletproof automation able to react to problems and manage subsystems without bothering anyone. Of course, even the best-run systems need housecleaning, which is where some extra automation is needed. IBM could build processes to FTP database backups and DBMS logs to an IBM "vault." A similar process might send Systems Management Facility (SMF) records, LOGREC and dump data sets to a site where IBM could analyze them for trouble. Another back-end process might do some rudimentary capacity planning and send helpful emails to the office IT guy that suggest appropriate upgrades and services.

IBM might find another revenue stream in offering periodic maintenance services. For example, once a month an IBM systems programmer could dial into the MM and perform whatever maintenance or configuration changes may be necessary. IBM could offer similar services for disaster recovery and database maintenance.

Why put z/OS on x86?
Knowing mainframes as they do, many people may not believe that IBM can create such a low-maintenance system. However, remember that the x86 mainframe is a stripped-down machine, running a couple of applications in a monoplex. The things that normally stress mainframes and require close watching, such as high resource utilization, cross-system signaling, complicated applications and disparate workloads, don't apply here. Instead, IBM builds the x86 mainframe as simply as possible to avoid the red line.

Other people would point out that the beauty of the mainframe is its ability to support a huge number of users running different workloads while driving hardware to very high levels of utilization. As you may have surmised, the x86 mainframe exploits none of these strengths and seems a little like using a jackhammer when a toothpick would do.

However, the x86 mainframe would give users the simplicity of distributed systems with the mainframe's reliability, availability, serviceability and security. In addition, because of z/OS' knack for supporting myriad workloads, customers will be able to add or consolidate applications without buying a new machine. Finally, when an office outgrows a box, IBM will be able to provide customers a clear migration path, up through larger x86 machines until they're finally ready to run on actual mainframe hardware.

But, again, the key to this whole process is cheap software. Running a mainframe on $1,000 of hardware won't mean much if it requires half a million dollars worth of software. That is up to IBM.

ABOUT THE AUTHOR: For 24 years, Robert Crawford has worked off and on as a CICS systems programmer. He is experienced in debugging and tuning applications and has written in COBOL, Assembler and C++ using VSAM, DLI and DB2.

What did you think of this feature? Write to SearchDataCenter.com's Matt Stansberry about your data center concerns at [email protected].

Dig Deeper on IBM system z and mainframe systems