More on “Slowgold”

See all articles from Sinc-Link v11 n4
Author(s): Hugh H. Howie
Publication: Sinc-Link
Pub Details: Sinc-Link v11 n4 (Volume 11, Issue 4)
Date: July/August 1993
Pages: 20

In the last issue I had some comments to make about SLOWGOLD, a program available from Dilwyn Jones Computing. This program is intended to slow the Gold Card down to enable some of the earlier QL programs to run on the now speeded up QL.

At that time, I mentioned that I had found a couple of “peaks” in the operation of SLOWGOLD, and although they were there, they did not in any way detract from the value of SLOWGOLD as a valuable asset to anyones library, as the amount of slow-down is adequate for most applications.

I have received further information from Dilwyn Jones as to his research of the problem, and his comments are so interesting, that I have decided to reproduce them here, in the hope they make more sense to you than they do to me. I just aint no technical man! If you want to completely understand what goes on, buy the program and try it out for your own self. You will NOT be disappointed.

QUOTE:-

We reckon by now that we know what causes the odd occurrences with some values for slowgold and it is very difficult to explain. It arises due to interactions between the length of the interrupt loop set up by Slowgold, the scheduler and other ‘regular’ items or interrupts in the QL. The values at which the effects occur varies depending what is happening 50 or 60 times a second. A small diagram may help to show what I mean, the – shows an interrupt or regular timed occurrence, where there is overlap there may be a clash or something is missed or queued, causing unpredictable effects to cycle round every now and then

Slowgold ---   ---   ---   ---   ---   ---   ---   ---   ---   ---

Scheduler
time slices  11112231111223111122311112231111223111122311112231111223

You can therefore, I hope, see that while Slowgold is in its own interrupt shown by the — above (the on time and off times are not in ratio, they will not be 50:50 as shown above) the normal actions of the QL do not occur and the time slices of the jobs currently active don’t occur as expected. Generally speaking, the bits chopped out of the time slices occur regularly and you get the same slowdown on a regular basis. But with some values the length of the interrupt will just happen to make things occur such that, for example, if job 1 is active, the interrupts could always occur when the scheduler is mostly attending to task 2 or 3 (even if they are suspended). I hope this is the real explanation for what is happening, we have suspected it for a while but never been able to conclusively prove it. Also, depending on the speed at which the machine is running, what else there is to consume time and so on, the slowgold interrupts may actually overrun their alloted time slots and strange effects such as apparently missed interrupts (leading to loss of slowdown) might theoretically occur, though we haven’t been able to prove this.