Hi Bryan,

On 24/10/14 15:40, Bryan K. Walton wrote:
> I've been troubleshooting a problem with large clock drift on my net6501
> for a while. It is running Debian 7, with kernel 3.2.58.
> 
> Of course, I've got a NTP server running on it.  And I generally have no
> problems with time sync during normal system usage (low load)
> 
> But what I've discovered is that during periods of high system load (like
> when I'm running rsnapshot on the box to do home backups, the clock starts
> drifting quickly.  The CPU load on the net6501 around that time is very
> high, around 2.5.  The clock drifts enough that ntp thinks that my time
> sources are failing sanity checks.
> 
> So then, NTP quits synchronizing the time.  After a day or two of this,
> with other high load times, the system time gets so far from the time of
> the outside time servers, that NTP just quits running.
> 
> Has anybody else noticed this?  Is there anything I can do to address the
> clock drift during high load times?  I never encountered this problem on
> my net4801.

we have suffered from the same issue. Applying the force-enable-hpet
patch [1] helps.

With that the kernel chooses tsc as clocksource and the system keeps
stable time with ntp.

Also, do not forget to pass intel_idle.max_cstate=1 to keep tsc stable.

HTH

Martin Sofaru

[1]
http://git.kernel.org/cgit/linux/kernel/git/tip/tip.git/commit/?id=2e151c70dfb0075ff83bec305c52a9da1ba49089

Attachment: smime.p7s
Description: S/MIME Cryptographic Signature

_______________________________________________
Soekris-tech mailing list
[email protected]
http://lists.soekris.com/mailman/listinfo/soekris-tech

Reply via email to