CA recently published and distributed a book titled Mainframe 2.0. The message can be summarized as: The mainframe has been around for 45 years; all it needs to prosper for the next 45 years is to support a new generation of administrators by providing administrative tools with a Windows/Linux look and feel.
While making the replacements for aging mainframe experts feel at home is definitely needed, it’s not a task that CA, IBM, etc. will fumble. Rather, the critical success factor for the mainframe immediately, and into the next decade, is figuring out how to encourage developers to create “virtualized” new programs and virtualized versions of existing programs.
Increasingly, with the advent of cloud computing, virtualization is coming to mean not just “virtual” in the sense of virtual machines and virtual memory, but also creating programs that run without change (binary compatibility) or at least without alteration in the source code via recompilation (source compatibility). The ideal of cloud computing is to provide the software infrastructure so multiple, related applications can be quickly moved en masse from existing, isolated data centers to an internal or external cloud (or, in EMC’s version, both) and allow them to be moved multiple times for workload balancing and adaptation to change. So far, this goal is far away. Programs based on Java Virtual Machines (JVMs) and Service-Oriented Architectures (SOAs) can theoretically be moved between Windows, Linux, and mainframe platforms, but, practically speaking, most users simply move from Windows to Windows, or Linux variant to Linux variant.
Now suppose Nirvana had arrived, and you could move massive applications between platforms daily (never mind about data stores; that’s a much more difficult problem). The obvious advantages are workload balancing for minimizing cost (just like “capacity on demand”) and lessened vendor dependence. A more subtle, but more profound, advantage is that applications or parts of applications can flow to the platform that specializes in their type of computing.
A simple way to categorize platforms is by their approach to scalability: scale-out by adding more machines (Wintel and Linux) vs. scale-up by adding more CPUs (high-end UNIX/Linux servers and the mainframe). The benefits of scale-out (throwing more legacy PCs at a problem for minimal cost) are well-known. Increasingly, scale-up (as exemplified by the mainframe) is showing advantages in the following areas: energy savings and carbon footprint; administrative costs and long-term, per-application Total Cost of Ownership (TCO); flexibility to handle wide application system-load swings; and, to some extent, security. This isn’t an either/or choice.
The wave of the future may be a new concept of “hub” in which computing of a particular type naturally gravitates to the best-suited platform, even when this means Web services are split across platforms. (IBM appears to be referring to the practice of assigning different workload types to different platforms as “hybrid computing” to avoid politically incorrect criticism of any of its product lines.)
An even subtler, longer-term advantage of this type of virtualization is “agility.” This is a carefully defined measure of relative improvements in a business’s ability to adapt to, and proactively embrace, change. For most modern enterprises, improvements in developer agility yield the biggest improvements in business agility—and therefore the biggest ongoing improvements in a company’s top and bottom lines—of just about any business strategy. The effort to virtualize existing software has a side benefit: It makes legacy applications set up as Web services usable for any development purpose, and therefore often dramatically speeds development compared to existing practices.
How does this relate to the mainframe? It means that, effectively, Wintel, Lintel, and the mainframe are in a race to see who can first virtualize their software. Who will be first to support moving the scale-up parts of Windows and Linux Web services to mainframe and UNIX server platforms? Who will be first to support migrating the scale-out parts of mainframe/high-end UNIX to Linux and Windows on other platforms?
Nobody is far along the road to software virtualization. If you, the user, want to move a z/OS application to Linux on the mainframe, there’s still some effort involved. If you want to move Linux to the mainframe, there’s now excellent support (as long as it’s Red Hat or Novell Linux) for the operating system and middleware superstructure for developing today’s composite applications and splitting them across platforms.
If you want to move Windows to the mainframe, we’ve just begun to see virtualization tools for new Windows/.NET applications, and we’re just beginning to identify ways of virtualizing existing Windows applications. If you want to move mainframe code to Windows, there are good ways to move COBOL, DB2, and CICS applications, while Assembler and ISAM or VSAM still require quite a bit of work by an expert. This means that, as internal and external cloud computing springs up, PC-heavy, external cloud server farms are preventing the likes of Amazon from taking full advantage of mainframe scale-up. In internal clouds, a cloud veneer is slapped on existing infrastructure that still can’t move from platform to platform.
So, to compete in this next decade—to ensure that customers see value in new applications on the mainframe—mainframe hardware and software vendors need to make the path to software virtualization exceptionally smooth for the developer. Then, external cloud suppliers need to respond by recognizing the advantages of the mainframe as a hub for certain types of customer needs and as a valid development environment. Users need to provide their own support for virtualization of their own software, mainframe or not. And non-mainframe platform vendors need to stop pretending that support for external clouds means their platform will win; they must get busy competing.
Folks, it’s all about the developer of virtualized services for clouds. None of these platforms are going away, but the platform that supports this developer first will win a greater share of new market, and should be the focal platform for mainframe and non-mainframe IT in the next decade and beyond. Start your development engines!