Parallel testing is the most accurate. Two teams enter current data into two separate computers, the old one and the one with the corrected code. The second team makes sure that the corrected code handles the data properly -- no shut-down. This testing may have to go on for weeks or months.
It will not be done. There is insufficient computer capacity to double the number of machines running code.
This posting on Peter de Jager's forum indicates that even if there were such excess capacity, it would not matter.
* * * * * * *
Date: Mon, 5 Jan 1998 13:59:28 -0500 (EST)
Date: Fri, 2 Jan 1998 21:21:11 -0700
From: email@example.com (Peter M. Ligotti)
Century Independent Parallel Testing: Possible or Not?
While reading about "parallel" testing, I had a realization that even if there was enough capacity, testing cannot even approach being parallel. Please bear with me as I express this thought and the steps of the logic. Please correct me if I have missed something.
Most large organizations intend to be ready by Jan. 1, 1999 to leave a year for testing.
Usually, parallel testing is to verify that the old and new system brings the same (hopefully) reliable results from the same inputs under the same conditions.
In other words, testing is normally century independent. Normally, testing can be done in one month or year to verify that the new system will run after testing the following month or the following year.
But this testing during 1999 will not give a clear indication that the system will run during 2000. In fact that is what we are testing for: Whether the system will run in 2000. Simulation is not enough. In order to test the systems capability to run in 2000, it HAS TO BE, actually be the year 2000. For example, if a date has been overlooked in the code somewhere in some system, that potential glitch might not show up in testing during 1999, but only show up in true perfomance after Jan. 1, 2000. Just one or two of these could sabatage an entire system, even infect other systems.
How could a parallel test ever be possible here? The old systems don't work after 2000, so what are we comparing with what? You have to have two straight lines for them to be parallel. Either we only have one line, or else they are not "straight."
We don't have two systems even theoretically such that both will work at the same time. Even the whole IDEA of parallel testing does not work here.
The only parallel test can be whether the new system(s) can work in 1999. But that is not what we need to know. We need to know if we have ONE SYSTEM that can work in 2000. There is no way to find this out before that date arrives. I think I may know the answer already.