NTP servers (Network Time Protocol) are an essential tool in the modern computer network. They control the time, ensuring every device on the network is synchronised.
Because of the importance of time in controlling nearly every aspect of computer networking accurate and synchronised time is essential which is why so many system administrators deploy a NTP time server.
These time servers use a single time source as a base to set all the clocks on a network to; the time is often got from the GPS network or radio signals broadcast from physics laboratories such as NPL in the UK (whose signal is broadcast from Cumbria).
Once this signal is received by the time server, the time protocol NTP then distributes it around the network – comparing the system clock of every device to the time reference and adjusting each device. By regularly assessing the drift of these devices and correcting for them NTP keeps clocks accurate to within milliseconds of the time signal and when this signal emanates from an atomic clock – it ensures the network is as accurate as physically possible, but what happens if you lose the time signal?
Damaged GPS antennas, maintenance of time signal transmitters or technical faults can lead to a NTP time sever failing to receive a time signal. Often, this is only temporary and normal service is resumed within a few hours but what happens if it doesn’t, and what is the effect of having a failed time signal?
Fortunately, NTP has back-up systems for just such an eventuality. If a time signal fails and there is no other source of time, NTP cleverly uses the average time from all the clocks on its network. So if some clocks have drifted a few milliseconds faster, and others a few milliseconds slower – then NTP takes the average of this drift ensuring that the time remains accurate for as long as possible.
Even if a signal has failed for several days – or even weeks – without knowledge of the system users, this does not mean the network will drift apart. NTP will still keep the entire network synchronised together, using the average drift, and while the longer the time signal remains off the les accurate the network will be it can still provide millisecond accuracy even after a few days of no time reference.