Code: Select all
while (1) {
t0 = gettime();
sleep(1);
t1 = gettime();
print t1-t0;
}Yes, sleep() is an imprecise beast, but it should not be sleeping 2-3 seconds on an otherwise idle system. What you'd normally expect is +/- 10ms perhaps at worst.
Known problem? Any known solutions?
(I'm kind of wondering how many other things this affects in a booted vbox Linux environment in general, it can't be limited to just my test script!)
An ideal solution would be to recompile the kernel due to a known offending kernel configuration wrt vbox.
Here's some actual code that measures sleep(1ms) into buckets:
Code: Select all
#!/usr/bin/perl
use Time::HiRes qw( nanosleep time );
$| = 1;
@c = (0,0,0, 0,0,0, 0,0,0, 0);
$max = 0;
$total = 0;
while (1) {
for ($x = 0; $x<1000; ++$x) {
$delta = time();
nanosleep(1000000);
$delta = int(1000*(time() - $delta));
if ($delta < 2.0) { ++$c[0]; }
elsif ($delta < 4.0) { ++$c[1]; }
elsif ($delta < 8.0) { ++$c[2]; }
elsif ($delta < 16.0) { ++$c[3]; }
elsif ($delta < 32.0) { ++$c[4]; }
elsif ($delta < 64.0) { ++$c[5]; }
elsif ($delta < 128.0) { ++$c[6]; }
elsif ($delta < 256.0) { ++$c[7]; }
elsif ($delta < 512.0) { ++$c[8]; }
else { ++$c[9]; }
if ($delta > $max) { $max = $delta; }
}
$total += $x;
print "$c[0] $c[1] $c[2] $c[3] $c[4] $c[5] $c[6] $c[7] $c[8] $c[9] max:$max total:$total\n";
}Code: Select all
<2ms <4 <8 <16 <32 <64 <128 <256 <512 >=512ms
92845 4856 783 576 824 42 28 28 18 0 max:368 total:100000Code: Select all
<2ms <4 <8 <16 <32 <64 <128 <256 <512 >=512ms
100000 0 0 0 0 0 0 0 0 0 max:1 total:100000
), 1 for each vCPU, and a lot more for everything else. The host OS then schedules these OS threads as it sees fit. Additionally, due to the nature of virtualization (keyword VM exit), the vCPU OS threads will typically only be running continuously for significantly less than 1 ms. Why am I telling you all this? To give you some basic background information and to show that on the low level, even an idle system is quite complex.
