problem

[Eddie Kohler]

Hey Tom,

> I manipulated DRIVER_TASKS_PER_ITER and for the smallest value (1) it gave the 
> best results, concerning variation in delays. But the high peaks still remain 
> the same. This does not have anything to do with my own elements, because 
> when I just use the classical elements the same phenomenon is observed. But 
> hey, we're still looking further :-)

Could you characterize the actual delay variation more precisely? I'd be
interested in looking at a graph...

Also, have you considered raising the kernel's timer-interrupt rate? The
default (1/100s) means that a CPU-intensive program can lock out every
other program, including one with higher priority, for 1/100s. If you raise
the HZ value, you can reduce that quantum.

Eddie