Machines in a distributed system have separate physical clocks. Those clocks do not advance at exactly the same rate and are not adjusted at exactly the same instant.

Clock skew is the difference between clock readings at one moment. Clock drift is the tendency for that difference to change as clocks run at slightly different rates.

Clock-synchronization protocols can reduce error by comparing clocks through a network, but message delay is variable. A node therefore obtains an estimate with uncertainty rather than perfect global time.

Wall-clock timestamps are useful for logging, expiry and human-facing time, but they should not be assumed to provide an exact ordering of events on different machines.

When correctness depends on causal order rather than physical time, distributed systems use logical ordering mechanisms instead of pretending all clocks form one exact clock.