Hi,

I am a mainframer who has finished remediation and testing on an IBM 370/Pascal pair of systems, and I want to know: Are there any other large co's out there who 'got it'? Who figured out in time that this was important, had to be dealt with, spent the money, got the right people and put together a spare mainframe as a Time Machine? Or am I overconfident in extrapolating 'because we did it, surely other companies would do it right'?

-- Roland Teigen (roland.teigen@lexis-nexis.com), July 22, 1999

Answers

Roland, I'm delighted with your post, because of course there are lessons to be learned from successes as well as failures. Could you tell us more about your effort - remediation approach(es), time frame (especially time left for end-to-end testing, my particular bugaboo), size of the software collection, team size, etc.

-- Peter Errington (petere@ricochet.net), July 22, 1999.

Roland, reading your description, I'm almost positive you used year-field expansion (vs. say windowing) but nail down that point for me if you would.

-- Peter Errington (petere@ricochet.net), July 22, 1999.

Approaches fairly straightforward: scanned code for every possible expression of date, time, year, yy, etc. Also for hits on existing calls to code that manipulated dates and times. Time frame--I can only speak for that part which I have seen, I know discussions and meetings were going on since at least early 1997, and the decision to purchase a mainframe went back at least that far. I was formally assigned to lead the effort for 2 systems Jan 4, 1998. We did 2 things that I think were helpful. We conceptually divided systems into pieces, and proceeded as follows: Step 1, go through our normal intensive process that we go through to allow code to be put into production, then subsequently, run that section through the rigors of two weeks in the Time Machine(testing dates from late 1999 through 2001). In this way, we were able to avoid the kinds of problems that 'deathmarchers' will encounter when they try to cram hundreds of changes, all at once, in a system, and both rush it into production and into one big 2000 test. That's how you introduce many bugs into production, and how do you know what change caused which bug...very ugly. We were done with 99% of coding into production as of March 99, and had a few bits left that went in May 99. Since then, we have been end-to-ending. In addition, other systems used our code in their own 2-week tests, thus much of it has been through numerous Time Machine slots. We were doing Time Machine testing in 1998. My team consisted of myself plus 5 other programmers, one of whom had just retired from 35 years at IBM...a crack shot. Size of assembler system is about 500 programs, of size between 1000 and 17000 lines of code. So maybe 3-5 MLOC. The pascal system, maybe 1-2 MLOC.

-- Roland Teigen (roland.teigen@lexis-nexis.com), July 22, 1999.

How strange that my question, based on reading your description that arrived as my e-mail copy, should appear in this thread ahead of your description. Greenspun, what the hell is going on? Anyway, I have another assumption based on reading your description: It sounds like the documentationof the programs must have been excellent. Please confirm this also.

-- Peter Errington (petere@errington.net), July 22, 1999.