North American Network Operators Group|
Date Prev | Date Next | Date Index | Thread Index | Author Index | Historical
Re: Clocking Sources (was NTp sources that work in a datacenter (was Re: Is latency equivalentto RTT?))
Some of you are telecom types who think "Stratum 1" is telco clocking SONET/SDH/T1/E1 kind of timing that's 10**-11 accurate but doesn't know or care what time of day it is. Stratum 3/Stratum 4 SONET accuracy is a couple orders of magnitude lower. And wander (<10Hz) matters, but jitter doesn't matter much. Others of you are computer types who think "Stratum 1" is "A good NTP time-of-day source", and 10**-6 accuracy is good enough to keep your billing records consistant and the machines in your data center all showing the same time, though it's nice to do better. Stratum 2, 3, 4, ... is just machines that are N hops away from Stratum 1, so they're likely to accumulate jitter and constant differences from asymmetry, but not to particularly have problems with wander except if their Stratum 1 wanders. It happens that GPS can do both, though the telco people use cesium clocks steered by GPS as their building timing sources, rather than using raw GPS, of course. If the whole GPS system dies, the telco systems will gradually drift unless they rebuild the timing networks that it mostly replaced, but good clocks help. The NTP system, on the other hand, will respond by doing something like a bunch of ISPs getting feeds from tick and tock and distributing them, and if it drifts a bit it's really no tragedy.