Re: timekeeping regression?

[Brad Spencer <brad%anduin.eldar.org@localhost>]

Thor Lancelot Simon <tls%panix.com@localhost> writes:

> On Fri, Jun 07, 2024 at 09:16:25PM -0700, Greg A. Woods wrote:
>> 
>> As a side note:  NetBSD can't keep time under VirtualBox (on macos), at
>> least not the way I was trying to run it.  However it works A-OK under
>> the macos hypervisor (x86 that is, I'm using UTM to configure and manage
>> it, and UTM uses QEMU for device emulation).  It's using ichlpcib0 as
>> its preferred timecounter (with TSC having been given a q of -100).
>
> Time slips badly enough under heavy load that it can't keep time while
> compiling NetBSD on GCP VMs, either, even with ntpd running.  It loses a
> minute or so per hour.  This is with NetBSD-10.
>
> I believe Google's hypervisor for GCP is KVM-like.


ntpd won't correct time if the drift is beyond 500ppm (that is, in a
linear manor.. see your drift file for details on what your local
situation is), if I remember correctly..  and it will also give up if
the spikes are too intense.  However, chronyd has bigger limits and may
be able to deal with the situation.  You may want to try chronyd and see
if it can cope.

I have had to use chronyd instead of ntpd on a couple of my Xen guests,
from time to time (although lately, only on one of them, as PVH guests
with official 10.0 could use ntpd) and I ended up having to use it at my
final $DAYJOB with Triton when the zone was a KVM (and probably BYHVE
based) full VM running under Triton (OpenSolarisish with KVM or BHYVE
bolted on the side).







-- 
Brad Spencer - brad%anduin.eldar.org@localhost - KC8VKS - http://anduin.eldar.org