The Dollars and (Common) Cents of Centralized Systems

By Lou Washington

(Back to article)

A smart person once said that “centralized systems tend to de-centralize, and de-centralized systems tend to centralize.” This is what we are witnessing in the IT world.

For years, the mainframe dominated the world. Then, the networked PC with server nodes evolved in various forms, and it became quite chic to talk about the mainframe and its assorted subsystems and support personnel as some kind of genetic throwbacks stuck in Conan Doyle’s “Lost World.”

The mainframe became the symbol of an information-systems tyranny that could only be broken via the complete independence of the end-user to maintain their own self-designed, self-engineered systems running on their own personal machines; a sort of siliconized version of the "Me Generation."

Enter the anti-mainframe — the networked PC, the servers down the hall, the evolved network held together by miles of cables, fiber optics and coax. The guy who supported your system worked for your department and likely never met your IT director, if you still had one.

And throughout all of this, the mantra of saving money was being chanted. “Cheaper than that boat-anchor mainframe” was the phrase I remember hearing constantly.

Cheaper Than the Mainframe … Really?

Like the hundreds of space heaters, refrigerators and lava lamps running in office cubicles, the cost of the hundreds of locally installed servers is hidden behind all of the various departmental budgets paying for the hardware, software, support, climate control, training, wiring and maintenance supporting these distributed systems.

“Ah ha!” you say. That’s as it should be, the expense is closer to the expenditure and therefore it more accurately reflects the actual need in terms of the money spent.

But consider the following:

What’s your capacity?

The average mainframe is running at approximately 85-to-90 percent capacity. The average server is running at about 20% capacity; almost the inverse of the amount utilized by the mainframe. What does this mean? It means the company is paying for huge amounts of server capacity that will likely never be needed.

How are you cooling your systems?

My company’s CEO once remarked that “without software, the only thing hardware can produce is heat.” So, what are you doing with the “heat” all those servers produce? Your mainframe system may well be cooled by water coming off your HVAC chillers. In the winter, the heat from the mainframe could be contributing to the maintenance of a tolerable ambient temperature in your building.

Now, think about all of those closets, file rooms, hallways and odd corners now housing servers. How are they being cooled? Do you see fans running for hours on end standing in front of wide-open closet doors exposing the servers within? Maybe you’re not cooling your servers; maybe you’re just losing the data maintained on the overheated box.

How much does that cooling cost?

The lost data, the hundreds of fans purring away, the crashed system resulting from overheating because someone forgot to turn the fan on — again, it’s not visible. It’s hidden behind the individual department or group budget or it’s spread out in the general overhead for the building, division or campus.

How much space are you devoting to servers?

Once you devote a closet to accommodating a server, it kind of limits what else you might be able to do with that floor space. Look around your enterprise, count up the computing power you have spread all over the place, factor it down to what you might actually need for each of the systems now running on their own dedicated server and then add up the amount of space each of those boxes takes up.

I think you will likely find a lot of wasted square footage.

Everything Old Is New Again

Many companies have started to pull servers back into a centralized environment where they can begin to find the answers to the above questions. Row after row of rack-mounted servers are going to take up much less square footage than the decentralized, localized architecture.

Suddenly the raised floor, the limited-access IT area, the climate-controlled, positive-air-pressure, halon- (or whatever they use today) protected room is back. Even the water pipe, the subject of such ridicule over the years, is now returning to cool the rack-mounted servers filling these newly consolidated operating centers.

All of this “mainframe” stuff is now part of the equation when you are costing out alternatives as far as platforms in your data center. Be very careful that you are indeed comparing fruit with fruit when you start adding up the MIPS required, MIPS delivered and total cost of ownership for a mainframe-centered environment versus a server farm.

I recently read a blog posting from a gentleman who had started down one road and ended up changing directions. His Intel-based data center was approaching $850,000; his finished mainframe facility cost him around $240,000.

I’m not saying this is going to be the result for every implementation. But, as we move away from the decentralized “chaos” of servers anywhere and everywhere, and the concept of a centralized facility is embraced again, be prepared to honestly evaluate the two architectures on their true merits, liabilities and real costs.

Lou Washington is the master of MIPS at software and services provider Cincom Systems. In his spare time, he’s also a senior business manager. He can be reached at lwashington@cincom.com. Nearly one-third of Cincom’s customers still use its products running in the mainframe environment.