Hi John, there isn't really much to remove, but nevertheless I followed your advice and replaced the sqlite select by a small standard tcl procedure. Maybe that set me on the right track.
There were also some exceptionally high execution times in between, but not in every test run and not always at the same repetition count. Those peaks had a slight tendency to occur at 500, 1000 and 5000 repetitions. From time to time they occured at other counts or not at all. I have a couple of daemons running on the machine, doing several things at more or less arbitrary moments. I suspect them to be the reason for this behaviour. One thing still boggles me: The peaks in my timing experiments with sqlite are reproducibly at a constant repetition count with approximately predictible times, as opposed the 'statistical' peaks in the sqlite-less case. Unless someone comes up with a similar timing description, I herewith declare this 'dilemma' solved. Thank you all for your help and sorry for the alarm. Kind regards Ulrich On Wednesday 22 February 2006 23:41, John Stanton wrote: > Ulrich, try designing an experiment which removes SQLITE and measures > the performance of the other software layers. That might resolve your > dilemma. > JS >
