Saturday, May 21, 2016

The Best Thing I Never Learned

The best thing I never learned was that I couldn't do something. It just doesn't occur to me that I can’t do something, particularly if it involves a computer. I’ve embarked on a number of projects over the years convinced that I would get them done one way or another, and a few of them ended up being pretty challenging. Which were, of course, also the ones I learned the most from. I also really hate to lose. Fueled by fearlessness to begin and a stubbornness not to give up, I share three of my fondest memories.

Gothic was the name of a computer program that would read a single input card and print it out on a old-style mainframe impact printer, transforming each character into many rows of asterisks that formed a large, gothic-font representation of the character. Those printers used boxes of paper that were one long continuous page, with perforations that allowed different printouts to be separated, so the Gothic program could print large, long banners, perfect for a “Happy 30th Birthday” sign. Gothic was written in Assembler language for a DOS (IBM, not Microsoft, the PC hadn’t been invented yet) operating system, but Wright State University used MVT and Gothic needed to be modified to run. I really wanted Gothic available, so I decided to figure out how modify it. The only problem was I didn’t know assembler language and it was totally Greek to me. Little by little I learned that a BALR was a Branch-And-Link-Register and that mainframes has 16, 32-bit registers. An MVC was a Move-Character and an S was a Subtract. And those translated exactly to machine instructions and that meant most of the program would work on MVT without change. That was a relief. I also figured out that there were only three things that had to change, the beginning of the program where control was passed from the operating system to the program, the end where control was passed back, and several operating system-specific inclusions which I learned were called macros. All those changes totaled perhaps a dozen lines of code, but a wealth of knowledge to me. And a bit of pride each time I saw a Gothic banner taped to a wall.

COBOL, an acronym for COmmon Business-Oriented Language, was one of the required programming languages to learn way back then, and Wright State used an interpretive version from the University of Waterloo named WATBOL, which was more efficient for student use. My final program of the quarter kept giving me errors and as much as I looked, I couldn’t spot what was wrong. One trusty way to debug a program was to insert PRINT statements at key places to verify the program was moving through its logic as expected. After much trial and error, I found that if I included a single PRINT statement at a specific spot, the error went away, which pointed to an error in WATBOL itself, not in my code. That’s a pretty hard conclusion to sell to your instructor, so I offered another solution. I would do all the work necessary to convert my program to real COBOL and convert all the input and output files from WATBOL format to regular operating system format, not really knowing if that was possible. And I would do it all in the week break between quarters. He agreed to delay my final grade for that week and another intense learning experience began. I had to learn enough JCL (Job Control Language) to create files and execute the COBOL compiler, find some utilities to do the file conversions and make the minor code changes due to the differences between WATBOL and COBOL. I was helped by having all the other students on break and no contention for computer resources. By the end of the week I had everything converted and successfully ran my program, without the PRINT statement, and turned in my completed assignment.

Years later, somewhere in the mid-90’s, I was taking a graduate-level Operations Research class where we were learning optimization techniques like Linear and Integer Programming and I needed an idea for a final project. At the same time I was managing the Network Systems group at The Mead Corporation, and they were in the process of acquiring a distribution business to add to their Mead Merchants group. My group had to plan and execute a networking solution for 30 new locations located all over the United States at the lowest cost while delivering good performance. Networks back then consisted of multi-drop lines, where you started a circuit at your data center, added a drop in an office in one city, added another drop in a second city, etc. We engaged AT&T to design a network for the new locations based on three rules. To deliver good performance, the expected busy-hour line utilization for each line could not exceed 50 percent (based on sales volume and line usage from existing Merchant locations) and a maximum of four cities per line. For high availability at the large regional offices, they had to be connected to two different lines. AT&T was given these specifications to feed into their world-class INOS design tool. And it occurred to me that this might make a good final school project. Took me a week to figure out that I would need to design an Integer Program consisting of 30 rows, each one representing an office and a large number of columns, each one representing any valid line given the performance constraints. The regional offices would just look like two locations with one-half the expected load each. The total distance from the data center to all the locations on a potential line would be used as an estimation of its cost. Add a few other constraints and solve for the lowest cost. In my head, it was possible, but a problem vastly too large to commute by hand, as we learned in class. But Mead had MPSX, a mainframe-based program for solving Integer Programs like this, so I learned enough about its syntax before writing two PL/1 programs to generate the matrix, which turned out to have over 20,000 columns. That completed, I submitted the batch job to run overnight before leaving for home, not knowing how long it might run. To my shock, it was still running the next morning. All this work and my final project was in danger. I cancelled the job and began trying to figure out what had happened. I noticed that most of the job’s activity was constant reading and writing to MPSX’s matrix disk datasets, so I changed the JCL to keep those datasets in memory and again submitted them for overnight execution, which it did, barely. I was at the edge of how large a problem MPSX could handle. Over the following week several changes were made and a final solution was given to AT&T to price and compare to INOS’s solution. I won by $100 per month, a small but decisive victory, and earned an A+ from my instructor.

Now that’s what I call having fun.

No comments: