[c-nsp] NTP synchronization problems C2801

Peter Rathlev peter at rathlev.dk
Thu Jul 1 17:23:00 EDT 2010


On Thu, 2010-07-01 at 00:10 +0200, Peter Rathlev wrote:
> I'd like it to run 15.0M, so the downgrade is just to see if it makes
> any difference.

I now tried to downgrade to 12.4(24)T3 on the previously somewhat
unsynchronized device. After reboot I cleared the NTP drift, deleted the
NTP config, reset the clock to 1993 and reinserted the NTP config.

The result was that the device was successfully synchronizing much
faster and much better than before. In a few momemts it ended up with an
offset of much less than 1 ms.

Strangely the original 12.4(24)T3 device couldn't go on keeping a sync
of ~1 usec; it now fluctuates up to 400 usecs. It still stays (well)
under 1 ms, so for most practical purposes it doesn't matter.

For the interested I graphed out the NTP offset of the two devices for
the last 30 hours:

The 15.0(1)M2 device, not synchronized too well:
http://img818.imageshack.us/img818/1673/c2801ios1501m2entservnt.png

The 12.4(24)T3 device, synchronized well:
http://img813.imageshack.us/img813/7313/c2801ios12424t3entbasen.png

(Graph times are GMT+2)

I'll let the downgraded device run for a day or two and then upgrade
again, waiting with excitement for the M3 release. :-)

The original problem was about micro precision IP SLAs measuring one way
jitter/latency in a switched network with 0.5-6.0 ms latency. It turns
out this is way more information than anyone wants, so we don't need the
ultra fine precision anyway.

-- 
Peter




More information about the cisco-nsp mailing list