Fewer timer interrupts when starting any user process

In node we saw that time is always drifting and the ntp jitter is very high. When we checked the number of interrupts via vmstat on the host and it was around 40-50 interrupts, which should usually be around 1000+ on these machines. When we stopped the java processes and checked for interrupts, they normalized around 1K. Also

cat /proc/interrupts ;  sleep 2 ; cat /proc/interrupts 


was showing about 200 interrupts when starting java processes and about 2k when processes were stopped.

I think timer interrupt delays might explain

  • High load on machines: since processes do not receive a processor kick after quanta, more processes are in the execution queue and therefore high load
  • Very slow answer: well, the commands we run may not get scheduled again due to the lack of timer interrupts after the quantum

But unable to explain

  • Low% cpu usage

There are so few questions here:

  • What happens to interrupts?
  • Timer interrupts have the highest priority (irq0) and cannot be ignored. So how does this happen (if at all) can the user leverage processes cause this?

source to share

1 answer

It looks like it's a hardware issue. The killing of the user program fixing the system was a coincidence. See also http://support.ntp.org/bin/view/Support/KnownOsIssues#Section_9.2.1.1 . talks about a similar problem.



All Articles