A validation methodology for fault-tolerant clock synchronization

Abstract

A validation method for the synchronization subsystem of a fault-tolerant computer system is presented. The high reliability requirement of flight crucial systems precludes the use of most traditional validation methods. The method presented utilizes formal design proof to uncover design and coding errors and experimentation to validate the assumptions of the design proof. The experimental method is described and illustrated by validating an experimental implementation of the Software Implemented Fault Tolerance (SIFT) clock synchronization algorithm. The design proof of the algorithm defines the maximum skew between any two nonfaulty clocks in the system in terms of theoretical upper bounds on certain system parameters. The quantile to which each parameter must be estimated is determined by a combinatorial analysis of the system reliability. The parameters are measured by direct and indirect means, and upper bounds are estimated. A nonparametric method based on an asymptotic property of the tail of a distribution is used to estimate the upper bound of a critical system parameter. Although the proof process is very costly, it is extremely valuable when validating the crucial synchronization subsystem.