Chapter 5

STEVE - February 1960

After studying the 1604 manuals for a while, and checking out a few journal articles, I became more convinced than ever that the analytic approach to the weapons systems analysis problem was not the right way to go. I wanted to use a Monte Carlo technique (which involves repeated use of random numbers and probabilities in a simulation model) and was convinced that it would be far more effective.

I set up a meeting with Dr. Edwards to discuss the project.

I started out by saying, "I think that rather than the analytic approach, a Monte Carlo simulation would be most effective here."

He interrupted impatiently "Wait a minute! What basis do you have for such a suggestion? Of course, I have a general idea as to what you're talking about, but haven't explored the details of the technique. I believe that the analytic approach is the right way to go."

"Let me gather some references and get back to you tomorrow. I studied this approach in a fair amount of detail last year, and I'd like to try to convince you that it will work well."

He grumbled and said "Well, OK - I'll listen to your proposal tomorrow, but we'll see."

He turned to some other papers and I knew that I had been dismissed. I knew I had seen some good articles in the Journal of the Association for Computing Machinery (JACM) and the IRE Computer group transactions. I also had my class notes and textbooks if I needed them.

That afternoon and the next morning, I rounded up and reviewed several key journal articles outlining the Monte Carlo process and its applications, and made copies. I also put together a one-page description of how I would organize the simulation.

When I went back in to my meeting with Dr. Edwards, I sensed that the good Doctor, although dubious, was thinking "maybe this will work, but more likely, maybe this will show this smart alec a thing or two." At any rate, he said "Well, I'll give you a chance to try this, but I hope you're ready to fall back to the other approach if this fails."

Over the next few days I started the design of the program and began with the code. I wanted to use the FORTRAN compiler that was available for the 1604 so that I didn't have to do all of the programming work in assembly language, although I was ready to do that if necessary. I also realized that my first challenge would be creating and programming the pseudo-random number generator subroutine.

I went back to the journals, and came up with a fairly efficient routine that used machine language fixed-point arithmetic and logical operations, which I proceeded to program in assembly language. I punched my program onto cards along with a 'wrapper' calling routine that would invoke the generator and output the results to the printer. I took my work to Isaac to see about running it on the 1604.

Isaac looked at the deck, didn't say much, but took it and went over to the console after putting the deck into the card reader. He loaded the assembler and let it process the input deck. Nothing came out except for a short error printout. Isaac pushed the form feed button to eject the listing from the printer, and brought it over to me. "Looks like you have a few assembler control cards missing," he told me.

"Where was I supposed to know about these?" Isaac just grinned and handed me a thin manual along with a few punched cards, saying "These should work for a simple program; put the first three before your program, and the last one at the end."

I combined the cards with my deck, and Isaac restarted the assembler program. This time, it churned for a few minutes, and then spat out a much longer error printout.

I took the printout and compared it to my coding sheets. I found that most of the errors were of my own making - bad instruction code mnemonics or bad address semantics, but a few stymied me and I needed to ask Isaac for help with a few error messages.

Isaac looked at my printout for a minute, and then broke into a big grin and laughed. He said "Well, those CDC guys didn't get the assembler quite right. There's a bug that isn't documented anywhere; I found out about it from one of their engineers. You need to work around some of the addressing modes that you were using which don't really work right in the assembler."

He then showed me how to modify my program to make it compliant with the assembler. I punched up the new cards, inserted them into my program, and we repeated the process. This time, the assembler completed with no errors and proceeded to punch an object program deck.

I pulled the object program deck out of the card machine punch hopper, and, before I went to put it into the card reader, I paused. I asked Isaac "Do I need some control cards or a loader here also?"

He gave me some cards which I combined with my object deck, and then put it into the reader. Isaac pressed the load button. The program deck loaded, and the printer started printing out the supposed random numbers, one to a line.

After about 10 pages, Isaac stopped the machine, asking me "Is this thing going to run all day? Did you program an infinite loop or will it quit after a certain point?"

Red-faced, I admitted that I had not programmed a limit. Furthermore, when I tore off the printout, I was dismayed to see that after a page of apparently random numbers, all subsequent output was of the form .00000000

Dejected, I turned to Isaac and said, "It's still not right. There's something wrong; the output looked good for a while, and then everything went to zero."

Isaac said, "Go back and check your algorithm. To me, it looks like once it generates a small value, it gets stuck."

Somewhat humiliated, I took all my listings back to my office. The next day, I figured out the problem with the algorithm and fixed my program. I also modified it to put the results out to tape rather than to the printer. Finally, I set it up so that it would stop after a specified number of iterations.

When I took the program back to Isaac, I told him "Isaac, I owe you a lot. You have graciously bailed me out from several beginner mistakes with this system. When I came in here, I thought you were just a maintenance tech; now I know better. You are a valuable resource on this machine, and I've learned a lot from you."

Isaac didn't say much, but his grin told me all I needed to know. We ran my revised program, and then loaded a statistical analysis program that could be used to analyze a sequence of numbers for randomness. The results of several tests indicated that my pseudo-random generator was going to be quite acceptable for the project.