Since 1992, I’ve regularly challenged the thinking of those who tout that the industry’s premier computing platform is doomed. As 1940s politician Wendell L. Willkie once stated, “I would rather lose in a cause that I know some day will triumph than to triumph in a cause that I know some day will fail.” The fundamental drawback of decentralizing a computing system is that it makes simple things more complicated. Conversely, the nature of re-centralizing systems makes complex things more simple. Furthermore, added complexity increases costs and points of failure while reducing scalability.
I’m still amazed at how many large companies set out to “kill the mainframe,” only to become “fashion victims” of the IT trend du jour. True, it’s possible to cherry-pick applications off the mainframe, but it’s a different game once you get past those easy pickings. I wouldn’t be a bit surprised if the failure rate of trying to move an interdependent set of mainframe applications off the platform approaches five nines (99.999%). Why? The mainframe routinely manages mixed workloads of transactions and batch jobs and e-business services at 90 to 100 percent CPU utilization without incurring failures. Typically, a Unix server will be sized to run at less than 65 percent average busy for fear of failures at peak utilization periods, and it seldom runs mixed workloads. Surround the mainframe with all the servers you like, but you can expect your costs to increase rather than decline. Major cost advantages of “centralized” computing “center” around resource sharing and associated economies of scale. Outsourcing firms have learned these lessons well; their profitability depends on how well they leverage re- centralization and server consolidation to keep system management costs low.
Server buyers beware! Peddlers of mainframe alternatives have often misrepresented the capabilities of their servers and their accomplishments. I once challenged a VP whose company sold high-end Unix servers because his CEO was making false public claims that their 11-year-old firm had “never lost a customer and never had a failed project.” I pointed out how one of their customers squandered roughly $300 million in a failed attempt to replace mainframes, to which he replied, “[My CEO’s] claims are true with two qualifications. First, we’ve never lost a customer in a market where we’ve remained, and second, we’ve never had a failed project where we didn’t try to warn the customer ahead of time.” Qualifications indeed! Not only did they dismiss a $300 million disaster, in reality this company also had abandoned an untold number of customers some years prior when they pulled out of the highly competitive, low-end Unix server market. Low integrity is at an all-time high among IT vendors.
Once started, IT myths easily perpetuate as most fashion victims have no plans to call a press conference to lay out the embarrassing details of their fiascoes. Nor should they! If you’ve just spent millions learning that something doesn’t work, keeping quiet avoids bad publicity and lets your competitors pay for their own education.
Some things may work on a small scale, but it’s a different ballgame in a large enterprise. Since 1992, IT planners have debated whether client/server is a hardware or a software architecture, when in fact it’s neither. It’s simply a high-level concept of how computers relate to one another. An architecture goes beyond nuances and notions; it provides the specifications for formats, layers, protocols, and processes.
Ironically, the most successful server-farm virtualizer to date has been the mainframe’s venerable Virtual Machine (VM) operating system, where virtual client machines access virtual server machines. One reason VM thrives is because the network between all the clients and all the servers is main memory, allowing inter-system communication to occur at memory speed, which is thousands of times faster than the fastest physical network connection. VM’s ability to support thousands of concurrent Linux images is also a nice plus. Otherwise, it costs a bundle to maintain server farms.
Another top-performing, server-farm virtualizer is the parallel sysplex, where a cluster of z/OS images are tied together with a hardware coupling facility. It took more than 20 patented innovations to create a coupling system that performed and scaled in a manner appropriate to the types and volumes of work mainframes process.
Though Josh Billings (1818-1885) was a humorist from another era, his quote rings true today for the IT industry: “The trouble with people is not that they don’t know but that they know so much that ain’t so.”
Grace and peace.