This year I turn 55, and I still enjoy playing competitive tennis in leagues and tournaments. Ten years ago, I was asked to play Jeff, the local high school 18-year-old phenom to help him prepare for an upcoming match against a perennial opponent whom he had never beaten. Someone watching us warm up would have likely noted Jeff ’s speed, power, and well-developed strokes. Watching me … well, never mind. Anyway, though I was 26 years older, I knew much more than Jeff did about resource utilization and strategy. Jeff also had a number of weaknesses I exploited, so I won. Afterward, we reviewed a few things, and a few weeks later he got past his nemesis at region.
My worst shellacking came in a practice match with former pro tour player, Russell Seymour, who is currently ranked seventh in the world in the men’s 75-age bracket! I played Russell when he was a spry 62 and tuning up for a national event (he has won several dozen national and international tennis championships during his career). Somehow I won the opening game, then he steamrolled me the next 18 straight games. Final score: 6-1,6-0, 6-0! Though Jeff, Russell, and I used similar equipment (shoes and rackets), we played the game at quite different levels. And though we all started playing at a young age, Russell’s development occurred at a much more advanced level. I’ve played many good players, but Russell is by far the best tennis player I’ve ever played.
In many ways IT can be compared to tennis, especially when looking at the different levels at which systems can perform work. Too often, IT planners overlook the operating system capabilities and track record, and limit their consideration to raw power and speed, which is a big mistake. For example, there are huge “skill-level” differences between Wintel, Unix, and z/OS. Just compare how each system handles mixed workloads, security, virtualization, cryptography, scalability, serialization, data sharing and clustering, and the differences become clear. Windows, like Jeff, has a number of weaknesses that are easily exploited. In many ways Unix is superior to Windows, but z/OS is by far the superior OS.
Mainframe pros know that oxen are better at pulling plows than using a thousand chickens. But will chickens ever close the gap? After all, smaller servers (the chickens) come with pretty impressive specs. They come in packages of two, four and eight SMP-CPUs, fit into a one-U rack- mount-format, 40 or more per rack. Some chickens even run HyperThreaded in dual, quad, and octa-way SMP- configurations, 64-bit, at more than 2.4 GHz per engine, have adopted virtualization in a useful, and affordable way and now are long-standing citizens in the data center. Plus, chickens are considered way cool because they handle Windows, Linux and Solaris 10, as well as the Java world, WebSphere, WebLogic, JBoss, Tomcat, Geronimo, and Web Services. With all of these attributes, can’t these chickens challenge oxen in size, performance, total capacity and cost, capital, and operation? Nope! While there’s plenty here to dazzle the server farm hands as they go about their chores, today chickens remain inferior to mainframes because they cost more to manage, don’t allow mixed workloads, aren’t bulletproof, don’t scale as well … yadda, yadda. For example, using a z/Series IFL (Integrated for Linux) specialty processor, one company’s sophisticated Java application experienced eight times more throughput than running on dedicated Unix servers. Since “distributed” technologies run dedicated workloads, they are notorious for low processor utilization. By contrast, even though the mainframe runs multiple workloads and allows data sharing, utilization regularly tops 90 percent.
Some would argue the chicken coop is looking more organized today, as blade centers and rack-mounted servers improve space and power considerations. Duh! This validates the mainframe approach that centralization has clear advantages for resource sharing, manageability, serviceability, and providing economies of scale. Mainframes can minimize the complexity of operation and disaster recovery.
I’m still amazed that mainframes are so sophisticated, yet so regularly maligned. IBM should make customers more aware of what sets the mainframe apart. Advanced design and decades of refinement explain why a mainframe can support thousands of concurrent users far better than the alleged alternatives. To limit platform comparisons to hardware speeds and feeds is like trying to pick the winner of a tennis match based on the equipment being used. Whether in tennis or computing, those who have superior capabilities and know how to use them have a huge competitive advantage!