>would not be surprised to find more areas in Linux where elapsed time is
>assumed to be equal to CPU time. ".

And how right he was :-)  We call it busy wait (as I understand
the way to do timing with finer granularity than 10 mS). In my
case 'iperf' computed the number of iterations of a tight loop
to do the timing, and then found it had to correct that (when it
lost the CPU for a while) so that it computed a negative number
of iterations, and thus stalled the process for a long time...

I can show you Linux virtual machines burning an entire CPU and
see vmstat and top tell you the system is close to idle (whenever
they get resources to produce the output for you). I think there
is a lot of work to be done.

Rob

Reply via email to