Monday, Mar. 31, 1975
All Clear?
To someone, somewhere--but whom?--the memorandum circulated recently at Harvard Business School's division of computer service doubtless made perfect sense. It read, in full: "We have been informed by DEC that a bug in the normalization algorithm used in three MACRO instructions (FADL, FSBL and FMPL) can cause a FORTRAN double precision compare to give incorrect results. A double precision compare should be accurate to 16 digits. This bug can cause the compare to give incorrect results in the ninth digit. We will notify all Users as soon as we receive a solution to the problem from DEC."
Ordinary mortals might blanch at such howlingly incomprehensible computerese, but the message turned out to be, if not altogether graspable, at least mildly approachable. The errant computer in question was built by Digital Equipment Corporation (DEC) and installed at Harvard five years ago, presumably reliably analyzing data at a steady, comforting clip. No one discovered until early this month that it was inaccurate in, of all places, the ninth digit --still quite serviceable for run-of-the-mill computer wizardry, but not the very best the machine was fully capable of. Some infinitesimally remote calculation was slipping ever so slightly out of its grasp. The flaw is so minuscule a problem that most routine users probably would not discover it unless they were weighing electrons, and the Harvard computer is due to be corrected shortly. What, in plain language, is the matter? "I could explain the whole problem to you," said an assistant manager at the business school, "but it would take several hours and a blackboard." Thanks anyway.
This file is automatically generated by a robot program, so viewer discretion is required.