Configuring the Time Synchronization Protocol (TSP)

Electing a new TSP master

To ensure that the time synchronization service provided is continuous and reliable in the face of unusual events, an election algorithm is available to elect a new master. Unusual events include:

the host running the current master time daemon crashes.
the master time daemon terminates because of a runtime error.
the network on which the master time daemon is running is partitioned or restructured.

This algorithm allows slaves to detect that a master has stopped functioning and to elect a new master from among the slaves. Because the failure of a master time daemon results in only a gradual divergence of clock times among the slaves, the election of a new master need not occur immediately.

Running timed on gateways that connect distinct Local Area Networks (LANs) requires particular care, because the time daemon may act as a ``submaster''. A submaster is a time daemon that functions as a slave on one of the networks the gateway is connected to and as a master on one or more of the remaining networks. Submasters are necessary to overcome the restriction of current transmission protocols' preventing broadcasts from being transmitted over multiple physical networks.

A submaster classifies the networks that the gateway is connected to into the following three types:

A slave network, which is a network on which the submaster acts as a slave. Only one of the networks can be a slave network.
Master networks, which are networks on which the submaster acts as a master.
Ignored networks, which are networks that are neither slave networks nor master networks.

The submaster tries periodically to become master on ignored networks, but gives up immediately once it detects that a master already exists on that network.