Small problems on large computers

Abstract

What kind of problem is it that, while best solved numerically, is too small to benefit from the cheaper, faster, more reliable computation performed by a large digital computer? The ONR-sponsored group working with the Whirlwind I computer at the MIT Digital Computer Laboratory believes that a problem too small for a large computer is a very small problem indeed, perhaps at best a one-of-a-kind computation involving only a few hundred multiplications. To support this contention, one must of course overcome several important difficulties. As an experiment toward this end, the Whirlwind computer has been made available without charge to at least thirty different groups including representatives of twelve of MIT's nineteen professional departments. In almost every case, the detailed programming and coding is being carried out entirely by the users with the occasional advice of members of the laboratory staff. This essentially solves the difficult problem of where one finds a staff large enough to keep a high-speed computer busy on small scale problems. But the success of the method depends on making the programming simple enough to learn and to carry out so that scientists and engineers can set up their own problems for the machine and still save time and money. What is required is an organization and a set of procedures chosen to simplify programming to the point at which a novice can quickly learn to code his own programs.