"Clock skew detected" and 127.127.1.0 in ntp.conf

Win32.Neshto · 02-28-2011, 05:26 AM

Hi!
I try to setup a HPC cluster with CentOS 5.5. But now there is no Internet connection available in the room where the hardware is located so I set up ntpd server using synchronization with local clock (maybe I do something wrong). Here is my ntp.conf file on a master node (the master node has IP 10.0.1.1, a file server has IP 10.0.1.2 and compute nodes are 10.0.1.3..10.0.1.11, comments are omitted):

Code:

restrict 127.0.0.1 
restrict -6 ::1
restrict 10.0.1.1 mask 255.255.255.255 notrust
restrict 10.0.1.0 mask 255.255.255.0 nomodify notrap

server  127.127.1.0     # local clock
fudge   127.127.1.0 stratum 8

driftfile /var/lib/ntp/drift
keys /etc/ntp/keys

Any time I try to install software building it from source when I run make I get a waring "Clock skew detected. Your build may be incomplete". Some software build successfully and run normally after install but some build tasks are failed for example when I try to build RPM's via 'make rpms' for BLCR I get an error message about invalid date in changelog. But the system date seems to be right. I suppose I have invalid settings of NTP. Can anybody help me?

TB0ne · 02-28-2011, 08:30 AM

Local clocks are, by default, stratum 16, and are untrusted time sources. You're going to need to get your time data from SOMEWHERE. Since there is no authoritative time source defined anywhere, your systems don't have a 'trusted' clock to sync to.

You could get a cheap USB connected GPS receiver, and get time from the GPS satellites. That would make your clock source a stratum 1 server, which is definitely trusted. Otherwise, you're going to have to figure out some way to get a network time source set up.

Valery Reznic · 03-01-2011, 07:13 AM

If I remember correctly, ntpd works only if initial time difference is small
So in additional to ntpd, you also need to run ntpdate on startup to get initial sync.

See manual or
http://en.wikipedia.org/wiki/Ntpd for more details

TB0ne · 03-01-2011, 10:02 AM

Quote:

Originally Posted by Valery Reznic

If I remember correctly, ntpd works only if initial time difference is small
So in additional to ntpd, you also need to run ntpdate on startup to get initial sync.

See manual or
http://en.wikipedia.org/wiki/Ntpd for more details

Indeed true. But, the ntpdate program is deprecated, and has been replaced in many distros with the sntp command.

Valery Reznic · 03-01-2011, 01:53 PM

Quote:

Originally Posted by TB0ne

Indeed true. But, the ntpdate program is deprecated, and has been replaced in many distros with the sntp command.

I did it about 10 years ago and even not on Linux, but on UnixWare - it's a Unix from the famous SCO.
While name of the programs to make initial sync changes, need to make such a sync is still here.

Anyway, thank you for the update.

TB0ne · 03-01-2011, 03:15 PM

Quote:

Originally Posted by Valery Reznic

I did it about 10 years ago and even not on Linux, but on UnixWare - it's a Unix from the famous SCO. While name of the programs to make initial sync changes, need to make such a sync is still here.

Anyway, thank you for the update.

No, you are quite correct, and it was a good catch on your part. NTP will not sync up if the clock is too far off.

If you try to run ntpdate on *MOST* current distros, they'll give you the message, saying the command is outdated, and give you the alternate. However, there may be *SOME* distros out there that still use ntpdate. The command is still very valid on older distros...works fine on some of my older RHEL servers, as well as my old SuSE systems too.