In the mid 1980s, a manager told me we needed a system to monitor NYSE trades for violation of trading rules. He gave me two weeks to investigate possible methods and come up with a estimate of what needed to be done and how best to do it.
The system I designed would come to be called the Depth Exception Monitor system. My design consisted of one module to be written by IBM and one by another programmer as well as installation of some new hardware. I was on the hook for seven modules. I also designed three test modules so I could check out my code without waiting for the install of the new hardware or for IBM and the other programmers to complete their pieces.
The system would initialize an in-memory ‘database’ from several existing files, process queues of in-bound and out-bound data, log all activity, take periodic checkpoints and do recovery as needed. It was designed to recover to point-of-failure in a less than a minute.
The Checkpoint code was ‘right down to the metal’, bypassing not only normal DASD Access Methods but even EXCP/XDAP processing. I wrote my own CCWs, including switching tracks and cylinders; did my own SIO/TESTIO. It locked the system every 5 minutes for less than 2 seconds (under VM – never timed it running native).
In three days, I designed the system, coded 10 modules (in Assembler language), keypunched the lot and assembled them. I ran a test using the three test modules to exercise my seven modules. All ten of the modules worked perfectly on the first try and never required changes.
It took IBM weeks to get their module working (and we had to hold their hand). The system went live in record time and worked flawlessly until the functionality was folded into another system under development.
Nota Bene: You need to be an experienced programmer to appreciate how epic this was. 😀