Cristian's algorithm works between a process P, and a time server S connected to a time reference source. Put simply:
If the RTT is actually split equally between request and response, the synchronisation is error-free. But due to unpredictable influences, this assumption is regularly not true. Longer RTTs indicate interference that is generally asymmetrical. Offset and jitter of the synchronisation are thus minimised by selecting suitable RTT from a set of many request/response pairs. Whether an RTT can be accepted at a given time depends on the drift of the clock and on the statistics of the RTT. These quantities can be measured in the course of synchronisation, which optimises the method by itself.
Cristian, Flaviu (1989), "Probabilistic clock synchronization" (PDF), Distributed Computing, 3 (3), Springer: 146–158, doi:10.1007/BF01784024, S2CID 3170166 /wiki/Flaviu_Cristian ↩