Re: [chrony-users] Fwd: Chrony malfunctioning at beaglebones

[ Thread Index | Date Index | More Archives ]

On Tue, Aug 25, 2015 at 9:27 AM, Miroslav Lichvar <mlichvar@xxxxxxxxxx> wrote:
> On Mon, Aug 24, 2015 at 11:24:38PM +0100, Nuno Gonçalves wrote:
>> I've already went ahead and set the machine at 6601, the one with the
>> kernel bug, with nohz=off:
>> debian@bbb1:~$ cat /proc/cmdline
>> console=ttyO0,115200n8 root=UUID=d601c7fe-2699-4e27-9580-7dccb64b521d
>> ro rootfstype=ext4 rootwait fixrtc coherent_pool=1M quiet nohz=off
>> So if you have a minute I would appreciate if you could also try with
>> this option as you mentioned.
>> I note that at least for the first few minutes there are no ugly
>> frequency numbers, so it looks nohz have effects.
> It seems disabling nohz doesn't fix this. With the maxslewrate setting
> and a good drift file chronyd stabilized, but when running without
> driftfile and maxslewrate the skew and frequency are insane again.
> Running adjtimex -t 9000 and observing the offset confirms the clock
> is not slowing down by 0.1s per second as it should.
> But I think it's good we ruled out the nohz mode as the cause, as that
> would probably be a nightmare to fix :). Hopefully it's something
> simpler.
> Thanks,
> --
> Miroslav Lichvar

Hi Miroslav,

Considering this previous test we did with the nohz cmdline option,
considering the bug outcome (that was related to nohz), and also
considering this other thread:!topic/comp.protocols.time.ntp/MKvQGqLbGqk

Do you believe there is some problem with the nohz cmdline option?

I'm asking this because It also doesn't seem to have any effect to me
with the beaglebone running linux 4.1.

Consider the two graphs in attach. I can see clearly a effect on PPS
performance when I run some program that wakes every 1ms, but setting
nohz=off doesn't have any effect.

I haven't tried to compile any custom kernel.


Attachment: nohz.png
Description: PNG image

Attachment: wake1ms.png
Description: PNG image

Mail converted by MHonArc 2.6.19+