The Mainframe Is Dead
by Chris Hunt

I know this to be true; the learned pundits who write articles in computer magazines all tell me so.

There is some disagreement as to why mainframes became extinct. Some experts feel that their failure to run Wordperfect for Windows caused their downfall, others that the world of Open Systems was closed to them. Or perhaps they all drowned in a primordial morass of COBOL ?

One expert tells us "Windows is the new standard. Everything else, including Unix, is doomed". Another that "What everyone wants is Unix, C, Open Systems and Client-Server". They both have a point, they cannot both be right, but all agree that the mainframe is dead.

We are asked to accept many uncomplimentary statements regarding the shortcomings of mainframe computers, and of ourselves, the people who use them.....

The cheapest way to run large numbers of small jobs is to use a number of personal computers or RISC based workstations. Existing mainframe applications, predominantly COBOL programs, are being moved to micro processor based computers. Excellent mainframe compatible COBOL compilers are available for DOS, OS/2 and Unix. Programs are used to translate mainframe data sets for use on the PC, using the procedure division of the COBOL program to determine what fields must be converted from EBCDIC to ASCII.

Download and recompile the code, download and translate the data, send the mainframe to the museum, and live happily ever after. With luck; virus free and hacker free, too !

Mainframe compatible COBOL compilers for the PC also have an EBCDIC directive which allows them to process untranslated mainframe data, and to preserve the EBCDIC collating sequence. Mainframe applications can be developed, tested and maintained on PCs, thus reducing the market for mainframes even in shops which intend to continue using them. PC based COBOL development and test environments are far richer in function and far more productive than their mainframe equivalents; this is also true of other languages such as Pascal and C, but these do not necessarily have mainframe compatibility or the EBCDIC directive.

CICS applications may also be migrated to, or developed and tested using, CICS products running under OS/2 or Unix. IMS applications may also be developed and tested on the PC.

New transaction processing systems, and old systems which must be rewritten, are being implemented less and less often on mainframes. The personal computer and the workstation are taking over the mainframe's traditional preserve. DOS, Windows and Unix are replacing proprietary mainframe operating systems.

Workstation power costs around $300 per MFLOP, PC power using the Pentium chip will cost around $50 per MIP. Mainframes cost upwards of $30,000 per MIP. Yes; a mainframe MIP may be worth more than a Pentium MIP or RISC MFLOP, but one hundred or more times as much ?

Software for microprocessor systems costs hundreds to thousands of dollars. Once. Software for mainframes costs thousands to tens of thousands of dollars per month. Every Month.

Another all pervasive problem with mainframe computers has been poor application programmer productivity, and the resulting application backlog. However much money is spent, end user satisfaction is just as far away. User departments are more and more declaring their independence and seeking solutions on smaller computers, using machines and people under their own control. Even when the IM department is a confirmed user of mainframes, new projects are frequently implemented on small computers by user departments. Perhaps the application backlog was once the mainframe programmers' job security; now it is their path to the unemployment line.

"Tell me what you want and I will tell you why it cannot be done" is not what corporations want, or can afford, to pay millions of dollars for. Small teams with a "can do" attitude, free from the bureaucracy of the IM department and under the control of the users, are providing required solutions quickly and cheaply.

The mainframe, especially when combined with the traditional IM department, is more expensive in terms of hardware costs, software licenses, and people costs. And it does not get the job done. What company would not wish to get away from it ?

Well, that is what we are being asked to believe.

So, the mainframe is dead. As dead as any other $30 billion a year business, or really dead? The truth is that mainframes are still selling, but volumes are down and their share of the total computer market is falling. This is not surprising, there are now cheaper and better ways of doing some things that were previously done on the mainframe. Economic recession has hit large corporations around the world, adding short term losses to what many expect to be a long term decline in mainframe sales.

One often hears one more, somewhat foolish, argument against the mainframe - that it is "old fashioned". So are C and Unix. Relational data bases, multiprogramming, multiprocessing, error correction code memory, RAID, virtual storage, REXX and most other high level languages, structured programming, and many other technologies first implemented on mainframes or minicomputers have found their way onto the supposedly more modern microprocessor based computers.

Contrary to the opinions expressed in the PC press; not every mainframe user is downsizing to the PC, and in fact the downsizing trend appears to be diminishing. Not all users of Open Systems and Unix have benefitted from the move away from the mainframe; lack of connectivity and the inability to expand the systems as the business grows have caused major problems, and portability is not always much more than a pious and unfulfilled hope. Some current implementations of Unix are almost as proprietary as is MVS. Some past users of Open Systems have reported great success, and major benefits, in moving back to the mainframe or AS/400. PCs and Unix are right for some companies and for some applications, and wrong for some others.

There are advantages and disadvantages to all current operating systems and hardware platforms. MS/DOS, OS/2, Windows, Unix, MVS and others all have a role to fill; as have mainframes, mini computers, PCs and workstations. No one of these is likely to be used for all future computing needs. The universal operating system has two things in common with the universal solvent; it does not exist, and if it did there would be nowhere to put it.

Many major new systems are still being implemented on mainframes, and many mainframe IM departments are efficient, and provide a good service to their users.

The mainframe may not be the best place to implement graphical user interfaces, or to run spreadsheets and word processing. It may not be the cheapest place to run small batch jobs, or to support some small to medium sized transaction processing systems. It is still the only viable platform for the truly enormous data base or transaction processing system. And, to use some rather silly terms which are currently in vogue, the mainframe arguably has the only industrial strength operating system for your mission critical applications; MVS. If there is another, it runs on the AS/400.

Since it is now so easy to interconnect systems, why not use the PC or workstation to interface with the user and to perform local application functions, and the mainframe as the file server for large data bases and for applications requiring very high message rates? With the ability to connect prodigious amounts of peripheral storage, to maintain enormous data transfer rates, offering unmatched security and data integrity, and with the ability to communicate with almost anything the computer industry has to offer, the mainframe has no peer.

IBM announced it's massively parallel RISC-based computer earlier this year. It has 8 to 64 processors providing up to 8,000 MFLOPS, and costs around $300 per MFLOP. This is a scientific and engineering system which runs AIX. IBM also talked about commercial versions, future machines with thousands of processors, very high bandwidth inter-processor data exchange, symmetrical multiprocessing, and the ESCON channel. IBM also have S/370 processor chips on expansion cards for micro channel machines. Will MVS run on massively parallel machines? I do not know, but it would make sense. In any case, the large massively parallel machine is a mainframe under any sensible definition of the term.

IBM has also announced that applications written to Open Systems standards will run under MVS. Even if the mainframes of today are not purchased in large numbers solely to run such applications, this does add to their value. IBM is a major player in Open Systems, and appears to be fully committed to participating in, implementing, and connecting to, the new systems architectures.

It is not the only way to perform serious computing, it's numbers may be on the decline, perhaps it is even about to be re-engineered to take advantage of micro processor technology, perhaps it will be used more as a giant file server than as an application platform; nevertheless, it is not always the most expensive option, for some jobs it is the only viable choice, and it is not about to disappear completely, so LONG LIVE THE MAINFRAME.

Colleagues of mine tell me that I am incapable of saying or writing anything about computers without mentioning ALCS. I almost made it.

Readers of this magazine will appreciate that mainframes also perform useful functions, and provide excellent performance, with software other than MVS. Since most of us have careers which are closely tied to mainframe computers, perhaps we should all wish IBM well in it's endeavors to gain continued acceptance of MVS and related products. After all, the TPF market place alone will never justify the continued development, support and manufacture of mainframe computer hardware.

Finally, the usual disclaimers. The opinions expressed here are my own, and do not necessarily represent those of my employers, Bedford Associates, Inc., or of IBM Corporation.