A statistical clock synchronization algorithm for anisotropic networks

A method to estimate the value of remote clocks in distributed systems is proposed. The method is able to deal with isotropic and anisotropic networks and includes a way to detect performance failures on single exchanges. It uses a statistical approach to estimate the relative drift of clocks and a round trip clock reading protocol to compute the offset. A good precision can be attained and maintained without exchanging too many specific extra messages. Numerical results obtained from a discrete event simulation are presented.<<ETX>>