Please login or register. December 10, 2018, 01:43:15 AM

Author Topic: XNTP and HACMP  (Read 12240 times)

0 Members and 1 Guest are viewing this topic.

fbergenh

  • Senior Member
  • ****
  • Posts: 40
  • Karma: +0/-0
XNTP and HACMP
« on: June 13, 2007, 02:03:01 PM »
Recently, we had some problems with an unstable HACMP cluster.

An engineer from IBM performed a Health Check and we made some changes according to his recommendations.

One thing he did not mentioned as a potential problem was the fact that the time on both nodes was not synchronized because of a bad xntp configuration. The difference between the two nodes was almost 10 secondes.

I fixed it myself, but I am not sure if he overlooked this issue or that a different systemtime on two HACMP nodes is not a problem at all.

Is there somebody who can give me an answer on this?

Thanks in advance.

Regards,
Frank


Michael

  • Administrator
  • Hero Member
  • *****
  • Posts: 1101
  • Karma: +0/-0
Re: XNTP and HACMP
« Reply #1 on: June 13, 2007, 06:50:23 PM »
imho - a 10 second difference is not an issue. The only place where time might be an issue is when a timestamp is generated for a volume group.

However, your application might not have liked a 10 second reversal of time.

In summary, it is good that you go it corrected, as I am only saying that 10 seconds is not likely to be an issue. Once you get above 30 seconds (opinion here, no hard facts !  8)) you are getting into a more risky situation.

One way - say the poor mans way - to sync time over several systems is to run the AIX command setclock - and choose a common reference. I used to do this in classrooms because students would, by design, change the clock settings (exercise with the date command) and sometimes the year or month would be incorrect.

flok

  • Full Member
  • ***
  • Posts: 16
  • Karma: +0/-0
  • www.vanheusden.com
    • my opensource AIX software
Re: XNTP and HACMP
« Reply #2 on: September 10, 2008, 02:16:12 PM »
In a correct setup you would choose at least 3 but preferably 4 upstream servers (3 to see if there's a bad chimer, 4 to recover from one bad chimer).
Choose the same 3/4/+ servers on each HACMP node so that they all have the same notion of time: maybe incorrect if something goes wrong, but at least the same time.

Do not put 'ntpdate somehost' in your crontab (or whatever solution not using the ntp daemon) as that may cause large time-jumps which can confuse processes (e.g. cron).

Consider using the reference ntp daemon as the one included in AIX is antique. The one in AIX 6.1 is for example version 3:
bash-3.00# ntpq -c ntpversion -c version
NTP version being claimed is 3
3.4y

while the current version of the reference ntp daemon (the one from http://www.ntp.org/) is 4.2.4p5 from 2008/08/17.
Amongst several important bugfixes (e.g. security related) newer versions sync faster etc.