Authors: James Aweya Delfin Y. Montuno, Michel Ouellette and Kent Felske

Complete Citation

* Aweya, J., Montuno, D. Y., Ouellette, M., and Felske, K. 2006. Clock synchronization using a linear process model. Int. J. Netw. Manag. 16, 1 (Jan. 2006), 3-28. DOI= http://dx.doi.org/10.1002/nem.583

Abstract

In this paper, we present a clock synchronization scheme based on a simple linear process model which describes the behaviors of clocks at a transmitter and a receiver. In the clock synchronization scheme, a transmitter sends explicit time indications or timestamps to a receiver, which uses them to synchronize its local clock to that of the transmitter. Here, it is assumed that there is no common network clock available to the transmitter and the receiver and, instead, the receiver relies on locking its clock to the arrival of the timestamps sent by the transmitter. The clock synchronization algorithm used by the receiver is based on a weighted least-squares criterion. Using this algorithm, the receiver observes and processes several consecutive clock samples (timestamps) to generate accurate timing signals. This algorithm is very efficient computationally, and requires the storage of only a small number of clock samples in order to generate accurate timing signals. Copyright © 2006 John Wiley & Sons, Ltd.

Annotations

This paper introduces a method to maintain clock sync between systems using a linear process model. Overall this paper investigates how to maintain clock synchronization between hosts when no common clock is available. They only use this to maintain synchronization, they do not use this to obtain delay measurements, only jitter. Basically, they introduce a live model for maintaining clock synchronization between two nodes in a network using a "Discounted Least Squares Model". This model requires no significant amount of storage of past values, however it takes about 100 to 200 seconds to stabilize with a frequency offset of only 100ppm -- for our laptops the frequency offset is about 2ppm (assuming a low amount of jitter the synchronization occurs in the order of seconds).

Note however, that this should not be used to measure delay, because this method ignores the delay between systems. If the delay inbtween systems increases for a period of time and then decreases, it can mess-up the synchronization. Overall, I think we need to discuss this in reference to our clock sync paper to determine if there is anything worth pursueing.

-- DavidSalyers - 18 Jul 2007