[Pool] Pool server monitoring 12ms offset

[Ryan Malayter]

On Tue, Jun 8, 2010 at 5:43 AM, Shoppa, Tim <tshoppa at wmata.com> wrote:
> MC0FRED writes:
>> with reference to the other NTP servers I am usually within +/- 2ms of them.
>> Any ideas why the pool server thinks I am ~12ms off?
>
> All your NTP times come from network connections. Network connections have latencies.
>
> Typical latencies on the global network range from microseconds (local) to milliseconds (not far away) to 100ms+ (farther away or skinnier pipe).
>
> NTP makes a simple assumption that the latency is symmetric, that the transmit path and receive path both have the same latencies.
>
> >From the consistent 12ms offset that the NTP pool server sees, there is some kind of asymmetric latency going on.
>
> Asymmetric latency can be related to assymetric signaling (e.g. ADSL) but you don't expect traffic to be so flat to give such a consistent 12ms delay.
>
> You might check locally, if you have any very complex routing or firewall rules that may in fact be adding this latency in one direction. It is unusual for the latency to be so consistently 12ms.
>
> NTP local clocks can also have intentional delays added (e.g. fudge flags in ntpd.conf) but this seems unlikely.

Or it could just be "the Internet". Asymmetric routes are the norm
when crossing ISP boundaries, not the exception. Everyone does "hot
potato" routing, and dumps packets onto their peer and upstream
networks at the nearest exit point from their own network.

For example, when a packet travels from my hosting provider in Dallas,
it jumps from the hosting provider to Level 3 to XO in the Dallas
metro area, and then XO transports it to Chicago over their network.
But a packet travelling from my Chicago office starts on XO, and is
dumped to Level 3's chicago peering point, and then travels to Dallas
via Level 3's network.
-- 
RPM