Timer problems on FreeBSD 11R on KVM

[Dmitry Marakasov]

* Allan Jude (allanjude at freebsd.org) wrote:

> > I'm running a bunch of FreeBSD VMs on VPS provider which uses KVM.
> > Recently one of them began bahaving strangely. Symptoms:
> > 
> > - Much ~(30%) CPU spent in intr
> > - top shows CPU used by "intr{swi4: clock (0)}"
> > - kernel: calcru: runtime went backwards from 13958 usec to 10792 usec for pid 919 (sshd)
> >   messages in logs
> > - There's no much load, no high network traffic, systat shows no excess
> >   interrupts.
> > - This intr load is sporadic in form of ~30 minute busts each several
> >   hours
> > - The box had ntpd running, but disabling it doesn't help
> > - Playing with timer-related sysctls does not help either
> > 
> > I've tried i8254 and TSC-low timecounters:
> > 
> > kern.timecounter.hardware: ACPI-fast
> > kern.timecounter.choice: TSC-low(800) i8254(0) ACPI-fast(900) dummy(-1000000)
> > 
> > and kern.eventtimer.periodic=1
> > 
> > to no effect.
> > 
> > 
> > I suspect that this began after some changes on the hoster's side.
> > 
> > - This VM was working for more than a year without problems before
> > - A graph which monitors a reply time of website running on that VM
> >   showed min. response time jump from 50 to 75 msec on Monday, after
> >   which problems began.
> > 
> > So while I'm getting the hoster to tell me what they have changed, any
> > ideas?
> > 
> 
> Look at:
> kern.eventtimer.choice
> 
> and try different kern.eventtimer.timer values

Thank you.

kern.eventtimer.choice: LAPIC(400) i8254(100) RTC(0)
kern.eventtimer.timer was LAPIC, trying i8254.

-- 
Dmitry Marakasov   .   55B5 0596 FF1E 8D84 5F56  9510 D35A 80DD F9D2 F77D
amdmi3 at amdmi3.ru  ..:  jabber: amdmi3 at jabber.ru      http://amdmi3.ru