Numerical methods, large codes, and programmers
Paul E. Saylor · ACM SIGNUM Newsletter · 1980
I asked Johnson if solving linear equations was a big part of the computational load at the laboratory. He said it must be about seventy-five percent of the machine time, but that his work to try to improve linear equation routines was mostly ineffectual. "You would go into the code and think you had it running faster, then they'd change the geometry or coefficients and that'd just kill you. You tamper with that code and then it might run a hundred hours of 7600 time over the weekend and not be worth a damn. It'd come back and be nothing but spaghetti. You get burned a few times like that and you think twice about changing anything. But like I say, if you could figure out a way to do it better it sure would make a difference. Only I never could."