Hello Paul,
I have a new GNU/Linux installation sharing an UML installation.
(User Mode Linux is a virtualisation technology that allows multiple
Linuxes to run on a single physical system.)
I'd like to run ntpd on this UML system.
After giving ntp.conf some pool servers to use, I noticed that the
time seemed to drift, and I started to get "time reset" /
"synchronisation lost" messages in my syslog. Also interesting is
that ntp.drift read 500 after one day's operation and did not change.
I am guessing there is a fight for syncronisation between the host and
guest.
There is a limit to the amount of correction that ntpd is willing to
apply on a clock. It looks like you hit that limit.
Having a look on Google, it seems that guest operating systems derive
their time from the UML host. This seems a bit odd, because you can
actually set the (software) clock using "date --set" in the guest OS.
(Perhaps the time is slewed back to the host's idea of time... I
wonder...)
I'ld guess that the UML guest stores the offset to the host time.
So I figured that if the host keeps the guest in timesync, I would
just have to set ntpd to use the local clock as a reference driver. I
inserted these lines into /etc/ntp.conf:
server 127.127.1.1
fudge 127.127.1.1 stratum 4
This seems to work, but I wonder if this is correct...
This is only correct when you are SURE that the local clock is properly
synchronized. Does the host run ntpd and can you be sure that it is
ALLWAYS stratum <= 4?
My questions are thus:
1) Any comments on the UML situation?
2) Assuming the host was guaranteed to keep good time, is using a
local refclock the appropriate solution?
If you are running ntpd on the host, why not add the host to
pool.ntp.org? More accurate time for the clients too, less latency.
3) is a UML setup acceptable as a pool server?
4) is anyone else doing something similar?
_______________________________________________
timekeepers mailing list
[email protected]
https://fortytwo.ch/mailman/cgi-bin/listinfo/timekeepers
