"slewalways yes" not recognized in /etc/ntp.conf file

[John Jimenez]

Earlier today I opened up the following thread

http://forums13.itrc.hp.com/service/forums/questionanswer.do?threadId=1384815&admit=109447627+1257546685679+28353475

It looked like adding these 2 lines would be okay on HP-UX 11.23 system.
slewalways yes
disable pll

But after stopping and starting xntpd it looks like the first one is being ignored. How do I enable slewalways?

/sbin/init.d/xntpd start
Nov 6 14:54:39 maxux3 xntpd[27954]: tickadj = 625, tick = 10000, tvu_maxslew = 61875
Nov 6 14:54:39 maxux3 xntpd[27954]: precision = 13 usec
Nov 6 14:54:39 maxux3 xntpd[27954]: configure: keyword "slewalways" unknown, line ignored

[Matti_Kurkela · Accepted solution]

"slewalways" is a new configuration option in ntpd 4.x versions. HP-UX 11.23 still uses an older version, which does not include this ntp.conf option.

But you can achieve the same thing by using the "-x" command-line option when starting xntpd. Edit /etc/rc.config.d/netdaemons and change the line:

export XNTPD_ARGS=

to:

export XNTPD_ARGS="-x"

If your system cannot stay within the normal slew-only limit of 128 ms in normal operation, you have a poor-quality (congested) network with unpredictable latency or something is wrong with your hardware or your timesource(s).

A common mistake is to configure two timesources that do not agree with each other: in this case, xntpd flip-flops between the two, adjusting time back and forth.

Either configure _one_ reliable timesource, or if you want redundancy, configure at least three so that any two can out-vote a malfunctioning one.

Please read the "slew" chapter in the xntpd man page:

http://docs.hp.com/en/B2355-60105/xntpd.1M.html#d0e464117

MK

[John Jimenez]

Hi Matti,

Thank you very much for you suggestion. I made modification as follows to /etc/rc.config.d/netdaemons and then stopped and started xntpd but it is still being ignored Here is syslog

Nov 9 07:40:10 maxux3 xntpd[9929]: tickadj = 625, tick = 10000, tvu_maxslew = 61875
Nov 9 07:40:10 maxux3 xntpd[9929]: precision = 8 usec
Nov 9 07:40:10 maxux3 xntpd[9929]: configure: keyword "slewalways" unknown, line ignored
Nov 9 07:45:00 maxux3 xntpd[9929]: synchronized to 10.0.8.1, stratum=2

here is /etc/rc.config.d/netdaemons

export NTPDATE_SERVER="10.0.8.1"
export XNTPD=1
# export XNTPD_ARGS=
# To allow slewing in version 3 change the line above to below on 2009-11-09 John J.
export XNTPD_ARGS="-x"

I also tried modifying my ntp.conf from version 3 to version 4. but that did not work

Nov 9 07:29:21 maxux3 xntpd[9162]: inappropriate version number 4, line ignored

I will read the link you gave me and see if that shed any insight. I am also going to research getting to version 4. Is it these HP clients that need version 4? or the Windows Domain Controller that needs to bump up to version 4

[Matti_Kurkela]

No amount of configuration can make the standard xntpd (version 3.x.x) of HP-UX 11.23 recognize the "slewalways" keyword, because that keyword was a new feature of ntpd 4.x.x series.

But the

export XNTPD_ARGS="-x"

should do the same thing as the "slewalways" keyword does, so you should remove the "slewalways" keyword.

For the same reason, the standard xntpd of HP-UX 11.23 cannot use NTP protocol version 4 at all. You'll just get this error message:

> Nov 9 07:29:21 maxux3 xntpd[9162]: inappropriate version number 4, line ignored

The jump from xntpd 3.x.x to ntpd 4.x.x was a big change: dropping the "x" from the front of the name was supposed to be a clue that something major has happened. You seem to be reading the documentation written for the newer ntpd 4.x.x and trying to apply it to the older xntpd 3.x.x. Of course the new features of ntpd 4.x.x are not available!

In the thread you referred in the original post, Doug Burton had a piece of Cisco hardware that required NTP protocol version 4. The only way to make that work with a HP-UX 11.23 would be to stop using the standard xntpd and build & install a custom ntpd 4.x on the system. If you do that, you're on your own regarding ntpd patches. But if Doug really needed NTP v4, he could have done it.

I haven't checked if the HP-UX 11.31 includes an updated ntpd 4.x.x instead of old xntpd 3.x.x; I hope it does.

MK

[John Jimenez]

Thank you very much. That makes sense. Is it okay to leave the other line below in?

disable pll

[Matti_Kurkela]

The "disable pll" option is supported by the standard xntpd of HP-UX 11.23... but does it really do what you want?

From the man page of xntpd:
http://docs.hp.com/en/B3921-60631/xntpd.1M.html#d0e549939

----
enable auth|bclient|monitor|pll|pps|stats

disable auth|bclient|monitor|pll|pps|stats

Provides a way to enable or disable various server options. Flags not mentioned are unaffected. Note that all of these flags can be controlled remotely using the xntpdc utility program. Each of these flags is described below.
[...]
pll

Enables the server to adjust its local clock, with default enable. If not set, the local clock free-runs at its intrinsic time and frequency offset. This flag is useful in case the local clock is controlled by some other device or protocol and NTP is used only to provide synchronization to other clients. In this case, the local clock driver is used. See the Reference Clock Drivers subsection for further information.
----

To me, it seems to say that with "disable pll" in effect, the local clock is _not adjusted at all_ and the only use for xntpd with pll disabled would be for serving time information received from other sources to clients over the network.

Apparently the advice to use those two configuration directives comes from Oracle Metalink document 759143.1. This is about a leap second problem, so the time when the problem may happen is very predictable: on 31-December or 30-June between 23:59:59 UTC and 00:00:00 UTC, and not at any other time.

Leap seconds are used to keep human timekeeping aligned with the minor deviations in the Earth's rotation, so it is added in either of those timeslots whenever astronomical observations indicate that it's necessary. See: http://en.wikipedia.org/wiki/Leap_second

The organization responsible for the coordination of leap seconds will announce any upcoming leap seconds about 6 months in advance.

On 4-July-2009 the IERS issued a bulletin stating that there will be *no* need for a leap second on 31-Dec-2009:
http://hpiers.obspm.fr/iers/bul/bulc/bulletinc.dat

Furthermore, the Metalink document seems to indicate that the problem is specific to Oracle versions 10.1.0.2 to 11.1.0.7 running on 64-bit Solaris hardware.

The handling of the leap second is very OS specific. Some OSs actually indicate 23:59:60 UTC when the leap second is applied, others, like HP-UX, will just "stretch" the timestamp of 23:59:59 UTC to cover two second's worth of actual time.

We have Oracle 10 databases running on HP-UX 11.23, have xntpd enabled, "disable pll" is _not_ in the ntp.conf file, and we had no problems on 31-December-2008, which was the last time the leap second was actually added.

I'm not a DBA and don't have access to Metalink. But based on the references to that Metalink article I could find with Google, I would assume that the advice is specific to Solaris and those particular Oracle versions and _not at all applicable_ to HP-UX.

MK