a conference sponsored by the NATO

This paper presents a Quantitative Method of estimating manpower resources for large programming jobs which represents a simple, feasible guideline. The method combines program structural data with historical data on the productivity of programming groups. This data is then adjusted to cover non-programming resource requirements. This is not a precise method and, in fact, it is not as good as an estimate based on sound experience. But it is offered as an aid to estimators working in unfamiliar areas and as a verification of estimates obtained by other methods. Introduction It is very common for the cost and schedule of large programming systems to exceed initial estimates. This is most often due to changes in the functional specifications but, even where the specification is fixed, it is difficult to estimate software resource requirements accurately. Not only does the main resource—manpower—vary widely in productivity and quality but the secondary resources, such as machine time and publications support, are frequently unavailable at the appropriate times. Most important, every large system is an aggregate of so many elements—program modules, organizational interactions, logistics, etc.—that no manager can precisely determine the amount of work to be done. On the other hand, managers of some large jobs, such as the Mercury-Gemini-Apollo mission support systems, have learned how to meet rigid schedules successfully. By examining the methods used in successful areas, it is possible to draft guidelines for use on new projects. 69 This approach has been taken by IBM’s Federal Systems Division (FSD). It draws on FSD experience in many large jobs since 1959 as well as on other relevant background. The result is a set of guidelines that, while subject to error, are useful for arriving at a quantitative estimate of manpower resources.The method is taught in the FSD Programming Project Management Course, where it has been recommended as an aid to estimating projected work (during proposal or early design stages) and as a cross check on estimates of work in progress. Among the more than four hundred managers who have taken the course were many experienced managers who corroborated the quantitative guidelines and few who contradicted them. Large systems dened For the purpose of this paper, a large programming system is a “structured aggregate of elements that satisfies a set of functional and performance specifications” and that requires • more than 25 programmers • more than 30,000 deliverable instructions • more than six months’ development time • more than one level of management. 7.2 ESTIMATING RESOURCES FOR LARGE SYSTEMS