The Share 709 System: A Cooperative Effort

• Authors: Donald L. Shell
• Venue: Journal of the ACM (JACM)
• Year: 1959

SHARE was created in August 1955. From the beginning, its aim has been to reduce redundant effort among various users of the IBM 704, insofar as the preparation of commonly used routines is concerned. The organization has operated on a cooperative basis—establishing a standard language of communication and exchanging a large number of programs which form the tools for effectively using the computer. This standard language of communication is that used in the SHARE Assembly Program, which was developed by United Aircraft Corporation and patterned after the first such program for the 704 which was developed under the writer's direction by the General Electric Company. This, in turn, was patterned to a considerable extent after the “Comprehensive System” which was pioneered at M. I. T. for use on the Whirlwind computer.At the SHARE meeting held in December 1956, the organization anticipated the forthcoming IBM 709 which was similar in a great many respects to the machine around which the organization was originally built. In order to have a set of programs available for use upon delivery of the new machine, the organization established a 709 System Committee which was given the task of preparing a “system” for using the machine. The writer was made Chairman of this Committee. Various installations volunteered individuals for service on this committee on a full time basis. The following people were selected by the Committee Chairman and the SHARE Executive Board for participation on the committee: Donald L. Shell General Electric Co., Cincinnati, O., Chairman Elaine M. Boehm IBM, New York Ira Boldt Douglas Aircraft Corp., Santa Monica Harvey Bratman Lockheed Aircraft Corp., Los Angeles Vincent DiGri IBM, New York Irwin D. Greenwald Rand Corporation, Santa Monica Maureen E. Kane IBM, Poughkeepsie Jane E. King General Electric Co., Schenectady Owen R. Mock North American Aviation, Los Angeles Stanley Poley Service Bureau Corp., New York Thomas B. Steel System Development Corp., Santa Monica Charles J. Swift Convair, San DiegoThe Committee's work led to the creation of a system which has been named SCAT (for SHARE Compiler, Assembler, Translater).The initial problem facing the committee was to define what was meant by a system. The scope of the anticipated system varied greatly in the minds of the various members of the committee. After due consideration, it was decided to specify a complete installation operational system with all the fundamental necessities for effective machine utilization supplied. The committee would use the best experience available and advance the art of machine usage, if possible. However, it would not attempt any features of such novelty as to endanger completion of the system in the time available. Some of the objectives sought were:simplicitynonstop machine operationcard-tape input interchangeabilityminimum programmer constraintsautomatic error detectionability to handle the SHARE 704 language, wherever possible. The most important advance made in the system over previous developments is in the area of program debugging. It was felt by the committee that if any true advancement were to be made in this direction, especially in higher language coding situations, it would be necessary to make the machine reply to the programmer in the same language that the programmer originally used to code his problem. The ability to perform this very desirable feat has not been accomplished in every instance. However, a rather long step has been taken in this direction.This step was made possible through the use of symbolic machine language, patterned closely after the SHARE 704 language, as the common denominator for all communication with the machine. Instructions written in symbolic form, as macro instructions,1 as algebraic statements, as subroutines or in any other acceptable form, are converted to symbolic machine language by the compiler and retained in an encoded symbolic form as the compiled version of the program. This encoded form of the program is called the SQUOZE deck. Thus, the symbolic machine language is used for compiler output. It is also used for communicating program changes and corrections at execution time and even for debugging output. This use of the symbolic language is, we feel, a major forward step in the techniques of machine utilization.A rather generalized picture of system usage is illustrated in figure 1. The men pictured there are actually the same person (the programmer) at different points in the system. Initially, at the upper left, he supplies his programs. These go to the compiler, C, in various formats—symbolic instructions, macro instructions, library subroutines, previously compiled SQUOZE decks, and perhaps ultimately FORTRAN statements. The compiler processes this material to produce a single SQUOZE deck which may now be passed along to the modify and load, L, portion of the system. The loading procedure accepts, along with the compiler output, corrections and modifications directly from the programmer in the same symbolic language used to write the original program.Whereas the compiler and loading procedures operate separately from the object program, the input, I, output, O, and debugging, D, systems are in essence attachments to the object program. These routines absorb the bulk of the burden of translation for the programmers. A set of fundamental routines are provided in a very flexible form—easily adapted to the specific needs of any particular problem.This whole procedure makes possible some very important advantages. First of all, one can make corrections in the same symbolic language in which the program was originally written. Thus, it is a simple matter to keep the program continuously up-to-date. Secondly, the machine has available at execution time sufficient information for the inverse translation—from absolute to symbolic machine language. Thirdly, it is possible to leave many program parameters undefined until execution time, which by former methods had to be defined at compile time. Further details as to methodology and procedure are outlined in the other papers of this set.Three methods were available for the development of this system:The 709 System Committee could assume complete responsibility for the system, including program coding and checkout.The Committee could completely specify the system and SHARE members would volunteer to produce component programs for the system.The Committee could specify the system, whereas coding and checkout would be done by the manufacturer. The planning effort expended by the committee amounted to about five man years. This, as can readily be seen, is a very large amount of planning time for such a system. A considerable number of alternative ideas were investigated and developed to some extent before being discussed and either discarded or finally adopted as part of the system. Since there were a number of rather divergent points of view represented in the committee, it took a great deal of effort to reconcile these approaches into a unified working whole.In addition to the Committee effort, the manufacturer has spent between ten and fifteen man years in order to implement the system. This time has been spent in detailed specifications analysis, programming, coding, checkout, and system integration.The completed system will contain about 46,000 words of instructions and tables. About 13,000 words of this total are needed for the compiler. The remainder is used for the modify and load program, the input and output translation systems, the debugging system and the supervisory control system. About 1,000 words of memory will be needed by the system at execution time. A considerable body of experience will be required before this effort can be properly evaluated. However, we shall here set forth a tentative evaluation. First of all, an excellent machine utilization system has been devised. It is a system which should be generally acceptable to all of the users of this particular machine. It has been recognized from the beginning that many local installations will wish to make changes and improvements in the system in order to adapt it especially to their needs. However, the fundamental procedures used throughout the system will undoubtedly be retained in every installation. In particular, the basic foundation, the encoded symbolic language, can and will be used as a common language from one installation to another and will serve as an excellent communication device.It is the writer's feeling that the planning took more effort than was really necessary to produce such a system. This was due in large measure to the enforced inefficiencies of committee action; particularly a committee which is supposed to reconcile many divergent points of view and which does not have a really effective decision procedure. Viewed in this light, it would appear that the very excellent talent which made up the Committee was not used in the most effective possible way. On the other hand, probably all the Committee members will agree that the final result is a better system than any one of the members would have developed completely on his own. Furthermore, it was obtained with a smaller expenditure of effort by any one installation than would have been necessary to develop any reasonable working system. Viewed in this light, it would appear that although these individuals may not have been operating in the most effective way, still the various installations which contributed these people are getting more for their money than they would have obtained by having them work at home for the same period of time. In addition, future communication of programs among the members of SHARE will be greatly enhanced and expedited through the use of this language which is common to all. These results have made the whole effort extremely worthwhile.