topic Re: NTP Parameter question in Operating System - HP-UX

NTP Parameter question

John Jimenez — Fri, 06 Nov 2009 16:48:39 GMT

Last week I opened up a thread on some issues with my Cluster The answers were very helpful. But now I have two more questions.

1) A few of my servers say
server version 3 prefer but another one just has the ip address /etc/ntp.conf file. Is there any difference? Will it know to use version 3? Is there newer versions out there?

2) I understand that I should be pointing to an outside NTP, not inhouse windows Domain Server. But Oracle had their own recommendation that resolved a server boot on clusterware. The fix that was posted fixed issues on Solaris, but it looks like it might apply to my HP-UX servers. They reccomended that I add the following in the ntp.conf. Does anyone know what this does or if its a good idea to insert on HP-UX 11.23?

slewalways yes
disable pll

Re: NTP Parameter question

Victor Fridyev — Fri, 06 Nov 2009 16:55:53 GMT

1. It looks like you can use both forms in the /etc/ntp.conf, but for HP-UX the format

server version 3

is preferable.

2. I use two servers in DMZ, which are pointed to servers according http://www.ntp.org/
Other servers use these two as NTP servers

HTH

Re: NTP Parameter question

Doug Burton — Fri, 06 Nov 2009 16:58:06 GMT

Newer versions and info can be found here: http://support.ntp.org/rss/releases.xml

However, I would suggest you get what is support from HP so head over to the software section to get it or gather up those dusty Application CD's/DVD's.

The version to use would matter as far as what is served up. If your NTP server is on v2 or 3 then your line would show that number.

We just saw a problem with some Cisco hardware where they NEEDED v4 so we pointed to an outside source (we're using v3).

Here's what I do (obviously this may vary)...

* Add the following to the end of /etc/ntp.conf file:

server 10.10.10.5 version 2 prefer # The NTP server
driftfile /etc/ntp.drift
statsdir /var/spool/ntp/
statistics loopstats clockstats peerstats
filegen peerstats file peerstats type day enable
filegen loopstats file loopstats type day enable
filegen clockstats file clockstats type day enable

Notice that the "statsdir /var/spool/ntp/" has a trailing slash.

* Add the stats directory:

mkdir /var/spool/ntp ; chmod 755 /var/spool/ntp

* Redirect NTP logging from the /var/adm/syslog/syslog.log file to another, easier to read file by adding the following in the /etc/rc.config.d/netdaemons file. Also add "1" so NTP will start at bootup:

export XNTP=1
export XNTPD_ARGS="-l /var/adm/syslog/ntp.log"

* Create the ntp.log and ntp.drift file:

touch /var/adm/syslog/ntp.log /etc/ntp.drift
chmod 644 /var/adm/syslog/ntp.log /etc/ntp.drift

* Stop (if it's already running) the XNTP process then restart it:

/sbin/init.d/xntpd stop ; /sbin/init.d/xntpd start

* Test your config by executing the following: ntpq -p

Re: NTP Parameter question

John Jimenez — Fri, 06 Nov 2009 17:02:46 GMT

Thank you both thats really good info. Any feed back about what oracle reccomended?

slewalways yes
disable pll

Re: NTP Parameter question

Michael Steele_2 — Fri, 06 Nov 2009 17:56:26 GMT

Hi

slewalways yes

If you turn this option on xntpd will try to always slew the time and
will never make large steps backwards or forwards.

disable pll

Without this option xntpd will not be able to correct time differences
greater than 128ms.

Slew is NTP error recovery. Slew is forward adjustments in time in ms, something like 128 ms.

Re: NTP Parameter question

John Jimenez — Fri, 06 Nov 2009 18:27:30 GMT

Hi Micheal,

It sounds like slewing is the way to go to keep my cluster up, if for some reason Time sync gets off and then it recognizes it later, it will change in such small increments that Cluster should stay up.... that is if both clusters change in the same increments.

But still not sure about disabling pll. If I disable the pll and time somehow gets out of sync by more then 128ms will it not even try to adjust?

It sounds to me like the slewing by its self should take care of it...Do you think?

Re: NTP Parameter question

Michael Steele_2 — Fri, 06 Nov 2009 18:55:23 GMT

Well, lets say your NTP server crashes and one of its many clients drifts more than 128 ms out of sync. That disable pll as well as the slealways yes arguement, (* they always have to be together I believe *), well be able to adjust the greater than 128 ms drift.

Re: NTP Parameter question

John Jimenez — Fri, 06 Nov 2009 18:57:03 GMT

Awesome. Sounds like this is just what we need on these two clustered servers. Thanks again for explaining this.

Re: NTP Parameter question

KSoward — Wed, 27 Jul 2011 19:20:57 GMT

We have had the same recommendation from Oracle on a clustered RAC setup in our environment .At first we were not willing to go with this setting but after reading a few threads we're warming up to the idea.

John, did this work for you?