Client/Server
References Case
April 25, 1993
For those not up to speed on high-tech developments, client/server computing is an attempt to reach beyond the host-based systems that are fast becoming old-fashioned. In these older systems, a central machine, usually a mainframe or minicomputer, performed all the computing tasks and sent the finished results to a user making a request via a terminal with little or no ability to perform the task itself.
But with client/server computing, complex computer tasks or applications-such as searches through payroll records-are subdivided into parts that can be handled by PCs, the "clients." The information the clients need to perform their tasks-information stored in complex data bases-is controlled and parceled out by the "server."
Early client/server hypsters liked to refer to this old style of computing as expensive computing. Their reasoning was that the cost of a unit of computing on a personal computer is a fraction of that on a mainframe or mini. Thus was born Myth Number One about client/server computing-that it is cheaper.
Today not even the vendors dare say that client/server computing is cheaper than the old method. Cambridge, Mass.based Forrester Research estimates that the costs of running a large PC network are some 300% higher than supporting a like number of users in a mainframe network. Gartner Group in Stamford, Conn., pegs the back-end costs per PC in a large-scale operation at $24,000 over five years. And Chapel Hill, N.C.- based Xephon, a market research firm, claims that even projecting to 1998, the cost per PC user will still be 505 higher than per mainframe user. That's $15,000 per PC user over a five-year period vs. $10,000 per mainframe user.
A host of anecdotal evidence supports this research. Although First Boston Corp. views its move to client/server as a competitive advantage that gives its traders faster access to data feeds, moving applications from IBM mainframes to workstations swelled the cost per user to $15,000 from $4,000. First Boston also found that many of the important jobs handled so fluidly in the mainframe environment, such as backing up prodigious amounts of data, were uncharted territory in the client/server world.
But-the hype goes on: It is said that client/server computing ]will free your company from that oppressive mainframe and those awful MIS people who seek to control company data almost as though that data were their own. Thus is born Myth Number Two, which states that networked PCs can replace mainframes.
It is true that a half-dozen different high-end PCs have the same raw computing power as any mainframe at a tiny fraction of the price. But a PC or a whole bunch of PCs networked together do not a system make, not when that system is conducting a critical computing mission such as that of an airline reservation system, or the trading system of an investment bank or brokerage. Backed by 30 years of development by thousands of vendors, such systems have a wealth of features that protect the data from intruders and ensure the integrity of the data stored in the systems' various computers.
Three years ago, managers at Microsoft Corp.-the apostle of client/server computing told editors of ComputerWorld that they would pull the plug on all their large systems within two years. Somehow that horizon has moved out another two years from today. If Microsoft could possibly pull the plug, it certainly would. But it cannot without doing itself a lot more harm than good.
Still the hype, oh, the hype. Once they are up and running, the song goes, client/server systems are just a whole lot easier to manage and operate than the dinosaurs they will replace. Computer people can easily transfer their skills to the client/server world and everything will be ducky. This is Myth Number Three.
The new-wave computing environments are at least an order of magnitude more complex than the proprietary systems they will eventually succeed. For example, it is possible when using Cobol, the lingua franca of the mainframe world, to write a single instruction and thereby execute a relatively complex task, such as a data storage management routine. Using one of the languages of the client/server world, like C++, the same programmer might have to write as many as a dozen instructions. This means that in addition to there being some $1 trillion of installed proprietary or legacy systems out there, there are tens of thousands of programmers and other personnel who need to be retrained. Retraining them is a monumental, time-consuming and costly investment. It is going on now at most Corporations, and it isn't going to be completed any time soon.
PacifiCare Health Systems Inc., a $2.2 billion HMO, found out the hard way through its rude introduction to a client/server system. I.earning as you go, as a technology executive there told ComputerWorld, gets really expensive and sets you up for many false starts. "A good deal of our client/server] costs went to paying for a learning curve that was much steeper than imagined," he said. Education ate Up the largest chunk of money PacifiCare has spent on client/server, with more than half of an estimated 200% cost overrun of initial spending projections going to training.
PCs will grow in power, in number, and in functionality and become the dominant platform for conducting critical computing tasks. Client/server computing will replace the old style of crunching numbers and processing data. Just bear in mind that there are serious consequences to rushing this process beyond the ability of the underlying technology to sustain it, or of the work force to maintain it.