At 6:26 PM -0500 1/27/07, Danny Mayer wrote:

>> I am migrating from Linux workstation to Mac mini (Intel 1.66 GHz). I
>> like most of this computer and this OS. The only thing is really
>> wrong: ntpd performance.

See .

--
Brad Knowles , Consultant & Author
Co-author of SAGE Booklet #15 "Internet Postmaster: Duties and
Responsibilities"
Founding Member and Platinum Individual Sponsor of LOPSA:

Papers: LinkedIn Profile:

_______________________________________________
questions mailing list
questions@lists.ntp.isc.org
https://lists.ntp.isc.org/mailman/listinfo/questions

Dmitry Ivanov wrote:
> Hello!
>
> I am migrating from Linux workstation to Mac mini (Intel 1.66 GHz). I
> like most of this computer and this OS. The only thing is really
> wrong: ntpd performance.
>
> Our three local NTP servers are stratum-2 and their offset usually
> does not exceed 2-3 ms.
>
> First, I have tried "official" way of using NTP on Mac OS X. This
> resulted in ~600 ms offset after 24h of operation.
>

It would help if you showed the state of things using "ntpq -p -c rv" so
we could get an idea what's going on.

> OK, I have created my own ntp.conf and restarted "Network Time"
> service. My old Linux box has the same configuration with three local
> servers.
>

What's "Network Time"

> After several hour of operation, Mac mini still has 64 sec. polling
> interval. At the same time Linux box has reached 1024 sec. Frequency
> displayed by "ntpq -c rv" is drifting between 6 and 8 all the time...
>

You shouldn't see a poll interval of 64 secs after that long so some is
definitely wrong but we can't tell without some data from the billboard.

> What is wrong with Mac mini clock? Do other Macs have the same
> instability?
>

It must be some kind of configuration issue. What version of ntpd are
you running? Also are your firewall's set up correctly? is 123/UDP open
to send and receive packets?

Danny
_______________________________________________
questions mailing list
questions@lists.ntp.isc.org
https://lists.ntp.isc.org/mailman/listinfo/questions

OK, right after reporting this problem, I've started to collect
loopstats on Mac mini. See graph attached.

It takes ~36 hours to achieve relatively stable frequency value after
reboot. I have seen values as low as 6.xx before started logging. My
servers stablize much faster than Mac mini. Is this Mac hardware
issue? I don't have other Intel Mac mini available and can't check it.

Now I see perfect polling interval of 1024 sec:

frukt:~ dimss$ ntpq -p -c rv
remote refid st t when poll reach delay
offset jitter
================================================== ======================
======
+twister.noc. 192.36.143.150 2 u 20 1024 377 7.590 0.073
0.332
*thunder.noc. 192.36.143.151 2 u 996 1024 377 5.595 -1.172
0.302
+www.solutions.l 193.190.230.65 2 u 994 1024 377 10.097
-0.495 0.153
status=0664 leap_none, sync_ntp, 6 events, event_peer/strat_chg,
version="ntpd 4.2.0@1.1161-r Fri Jan 13 11:36:23 PST 2006 (1)",
processor="i386", system="Darwin/8.8.1", leap=00, stratum=3,
precision=-20, rootdelay=37.959, rootdispersion=70.441, peer=28293,
refid=10.100.10.111,
reftime=c96845a0.8e640cfd Mon, Jan 29 2007 12:06:24.556, poll=10,
clock=c9684984.98d47b08 Mon, Jan 29 2007 12:23:00.596, state=4,
offset=-0.680, frequency=11.181, jitter=1.283, stability=0.049
frukt:~ dimss$


"Network Time" is Apple name for ntpd


_______________________________________________
questions mailing list
questions@lists.ntp.isc.org
https://lists.ntp.isc.org/mailman/listinfo/questions

At 12:34 PM +0200 1/29/07, Dmitry Ivanov wrote:

> It takes ~36 hours to achieve relatively stable frequency value after
> reboot. I have seen values as low as 6.xx before started logging. My
> servers stablize much faster than Mac mini. Is this Mac hardware issue?
> I don't have other Intel Mac mini available and can't check it.

Dunno. All I know about the subject can be found in the archives of
the list, and in the FAQ entry I referenced. I don't have any Mac
Mini hardware myself, at least not yet.

--
Brad Knowles , Consultant & Author
Co-author of SAGE Booklet #15 "Internet Postmaster: Duties and
Responsibilities"
Founding Member and Platinum Individual Sponsor of LOPSA:

Papers: LinkedIn Profile:

_______________________________________________
questions mailing list
questions@lists.ntp.isc.org
https://lists.ntp.isc.org/mailman/listinfo/questions

Dmitry Ivanov wrote:

> OK, I have created my own ntp.conf and restarted "Network Time"
> service.

Perhaps I'm asking the obvious, but does ntp.conf still contain what you
put into it? OSX tends to recreate the file according to its own ideas
each time you access the the time dialog.

I was forced to make the file immutable in order to make sure ntp kept
doing what I wanted. Perhaps there are more reasonable ways but I
haven't found any.

N

On 1 Feb 2007, at 00:52, Nero Imhard wrote:

> Dmitry Ivanov wrote:
>
>> OK, I have created my own ntp.conf and restarted "Network Time"
>> service.
>
> Perhaps I'm asking the obvious, but does ntp.conf still contain
> what you
> put into it? OSX tends to recreate the file according to its own ideas
> each time you access the the time dialog.

I have created my own /etc/ntp-my.conf and changed startup script /
System/Library/StartupItems/NetworkTime/NetworkTime to use this file.
BTW Apple uses /var/run/ntp.drift file to store drift value. This is
stupid because contents of /var/run is lost after reboot. I've
changed this to /etc/ntp.drift.

I am mostly disappointed by Mac. (Almost) perfect visual design, lack
of engineering. Misunderstanding of NTP is very clear but not the
only example of this.

>
> I was forced to make the file immutable in order to make sure ntp kept
> doing what I wanted. Perhaps there are more reasonable ways but I
> haven't found any.

Thank you for this advice! Looks like this is the best possible way.

Could all this information be added to FAQ?..

_______________________________________________
questions mailing list
questions@lists.ntp.isc.org
https://lists.ntp.isc.org/mailman/listinfo/questions

On 2007-02-01, Dmitry Ivanov wrote:

> BTW Apple uses /var/run/ntp.drift file to store drift value. This is
> stupid because contents of /var/run is lost after reboot. I've
> changed this to /etc/ntp.drift.

/etc/ is for configuration information; daemons should not be writing
there. /var/lib/ntp/ntp.drift would be a better place for the drift.file

> Thank you for this advice! Looks like this is the best possible way.
>
> Could all this information be added to FAQ?..

You can easily add this information to the Community Supported
Documentation at http://ntp.isc.org/support

--
Steve Kostecke
NTP Public Services Project - http://ntp.isc.org/

Steve Kostecke wrote:
> On 2007-02-01, Dmitry Ivanov wrote:
>
>
>>BTW Apple uses /var/run/ntp.drift file to store drift value. This is
>>stupid because contents of /var/run is lost after reboot. I've
>>changed this to /etc/ntp.drift.

is the /var tree realy trashed at boottime?

the idea with "/var" was to have a distinct path that needs RW access
were all other stuff from / , /bin/ ... can be mounted RO

For volatile storage /tmp and/or /var/tmp is used.

This would mean that apple is happily doing footperforation.

G!
uwe

On 1 Feb 2007, at 15:42, Uwe Klein wrote:

> Steve Kostecke wrote:
>> On 2007-02-01, Dmitry Ivanov wrote:
>>
>>
>>> BTW Apple uses /var/run/ntp.drift file to store drift value. This is
>>> stupid because contents of /var/run is lost after reboot. I've
>>> changed this to /etc/ntp.drift.
>
> is the /var tree realy trashed at boottime?

No. Only /var/run. Some linux distros do this as well. This is
correct in most cases. But writing ntp.drift here is stupid idea.

_______________________________________________
questions mailing list
questions@lists.ntp.isc.org
https://lists.ntp.isc.org/mailman/listinfo/questions