Interval-based clock synchronization with optimal precision

Abstract

We present description and analysis of a novel optimal precision clock synchronization algorithm (OP), which takes care of both precision and accuracy with respect to external time. It relies upon the generic interval-based algorithm of Schmid and Schossmaier [Real-Time Syst. 12 (2) (1997) 173] and utilizes a convergence function based on the orthogonal accuracy algorithm of Schmid [Chicago J. Theor. Comput. Sci. 3 (2000) 3]. As far as precision is concerned, we show that OP achieves optimal worst case precision, optimal maximum clock adjustment, and optimal rate, as does the algorithm of Fetzer and Cristian [Proceedings 10th Annual IEEE Conference on Computer Assurance, Gaithersburg, MD, 1995]. However, relying upon a perception-based hybrid fault model and a fairly realistic system model, our results are valid for a wide variety of node and link faults and apply to very high-precision applications as well: Impairments due to clock granularity and discrete rate adjustment cannot be ignored here anymore. Our accuracy analysis focuses on the nodes’ local accuracy interval, which provides the atop running application with an on-line bound on the current deviation from external time. We show that this bound could get larger than twice the necessary lower bound (“traditional accuracy”), hence OP is considerably suboptimal in this respect.