select(2) timeout precision

Viktor Vasilev viktor.vasilev at
Tue Nov 1 14:42:31 PST 2005

Hello fellow hackers,

I'm writing a piece of code, that basically does the following in a

  tick.tv_sec = 0;
  tick.tv_usec = 10000;
  gettimeofday(&t1, NULL);
  select(0, NULL, NULL, NULL, &tick);
  gettimeofday(&t2, NULL);

With FreeBSD 5.4-RELEASE I almost constantly get ~20000 microseconds
delta. That is with 100HZ kernel on PIII 500MHz or Sempron 64 2800+
running in i386 mode. It also doesn't matter if I use usleep or
select.  I've tested the code on linux with 2.6.11 kernel and there I
not only get the 10000 us sleep, but I can also set it to 5000 us and
it still runs very accurately.

So my question is, is this a FreeBSD limitation of some kind, and is
there a way to circumvent it?

Best regard,
I think and think for months and years. Ninety nine times, the
conclusion is false. The hundredth time I am right.
                               -- Albert Einstein

More information about the freebsd-hackers mailing list