USAFRet :
The Space Shuttle software never went through any changes? Never updated?
You could not be more wrong
They also changed hardware along with the software.
They changed the software to fix bugs and add capabilities to the shuttle. But they never changed away from the 80386 processor or to new versions of MS-DOS.
13thmonkey :
so you picked XP for a 15 year experiment knowing these facts:
1. there have been several OS's in the decade preceding this point.
2. hardware has been steadily increasing in power for the previous decade, there's even a law named after it
3. hardware is not designed to run for 15-20 years guaranteed without failure
3.1 you might get lucky and it does, do all of the voltages coming off that PSU match what they started as? (hint they won't)
3.2 because of point 2 a hardware failure 10 years into the experiment would result in you dumpster diving for spares, which would have unknown provenance.
3.3 that hardware would not in actuality be stable over time, reference voltages will shift as resistors and capacitors age.
#3 means nobody can do a proper study longer than about 5 years.
I didn't get to choose. The IT department chose the computers, and purchasing always puts every lab equipment purchase out on bids. State law required these.
The point is that we were able to do this with MS-DOS. We were able to start an experiment in 1985 and continue it until 2000. As long as IBM-PC type machines with ISA buses were available that ran MS-DOS, the drivers that came with the scientific equipment worked, and we could change out bad computers without changes in timing or data accuracy because the new computers worked the same way. We DID run the parallel-run process and verified no change in operation.
We went right from Windows 3.1 to XP because that's when the last of the PC-based ISA computers failed. We did not expect all of these problems. We got them anyway.
As soon as Windows 2000 became the dominant operating system, none of this compatibility worked anymore:
- The ISA bus was gone. The old lab equipment guaranteed to be accurate for 30 years (precision parts and NO electrolytic capacitors) would not work at all on new computers.
- Every time a new version of Windows was installed, the accuracy of the magnitude of the data collected by the scientific equipment was the same. What changed was the timing between data collections. We got totally different results when doing the parallel-run process. The equipment company said they had to make new interface hardware and software to fix the problem, and to hold on to the old operating system for about 3 years until they could have it ready and tested. Some makes went out of business or removed computer control because they could not keep up with Microsoft.
- Often a Windows update DOES change the system timing enough to make the parallel-run process fail. And at the rate Windows get updates now, another update comes in before the parallel run is finished.
Ralston18 :
Curious about one thing:
Are the "ceteris paribus rules" truly that rigorous?
I do understand the experimental need to maintain constants and change only one thing at a time.
The trouble was not the accuracy of the collected magnitudes. The problem was controlling WHEN the data collections happened and the times between them. We were getting missed collections - the next data overwrote the port before the computer got the previous data. Also, we had serious latency problems when the collected data showed a need for the computer to adjust experiment parameters.
We traced this to Windows having control of the computer (instead of the collection program) when these things happened.
When we went through the parallel run process, the comparisons failed. The new system either changed the timing between data collections or the timings between collections varied. There were missed collections. And often it totally missed the onset of the event being observed.
New computational tools were an issue only when the Pentium bug appeared.
We weren't forcing the use of old computer equipment, We couldn't get the new computers to work with the old lab equipment (which still had that 30-year warranty, so we couldn't justify the money for new while we had it). And when we tried to find new equipment able to do our experiments, either the companies had stopped making it, they stopped making it computer controlled, there was a 3-year or so wait to get it for the new OS, they told us to use MS-DOS, they went to timestamps, or they could not simultaneously meet our collection speed, timing accuracy, parameter write, and onset specifications that the old systems were able to do.
One company claimed it could do everything we wanted in experiment control with a Linux based system. We bought it for $5000. We could not make it do all of the things needed in the same data collection setup. It could do the speed we needed, but only if we did not look for onset or adjust the experiment parameters. If we set it up for the latter, we could not get the speed or the timing accuracy between collections. I was about to pack it up to return it when the lab manager got so mad he smashed it with a sledgehammer.
I finally found out that multitasking was causing out problems. MS-DOS worked because it didn't multitask unless you spooled a printout. The problem is that nobody sells a system that doesn't multitask. We can run MS-DOS on new computers, but the timing is off and we can't use anything but floppy disks.
Facebook
Twitter