Duncan Booth <[EMAIL PROTECTED]> writes: > [ ... ] > > If the context isn't switching enough for you then try calling > sys.setcheckinterval(n) with varying values of n until you find one which > is suitable. Calling it with a lower value of n will increase the frequency > that you switch thread contexts, although of course it will also increase > the overall runtime for your program.
Thank you very much. The sys.setcheckinterval function is what I need. It seems that the original writer of the app had set this interval to a high value in a part of the code that I overlooked until you mentioned this right now. > [ ... ] > > Why does it matter whether individual threads are being 'starved'? Surely > you want them all to complete in any case, so does it matter if they run > sequentially or in parallel? Because some of the threads perform monitoring and notification that need to occur in a timely fashion. Since these threads are doing IO, they switch context appropriately, but once one of the big number-crunching threads gets control, it starves out the monitoring threads, which is not a good thing for my app ... or at least it did so with the original large checkinterval. -- Lloyd Zusman [EMAIL PROTECTED] God bless you. -- http://mail.python.org/mailman/listinfo/python-list
