Experimental evaluation of simplified verification methods for responsive communication protocols

Abstract

We consider responsiveness verification of communication protocols on the client-server system CS/sub N/, which consists of one client site and N (N/spl ges/2) server sites, under the assumption that a single transient failure occurs on any one server site. For the verification, a typical naive method NAIV generates exhaustively all abnormal system states reachable after the failure, and then calculates the recovery time needed for returning to a normal state. We propose two simplified methods (RSY M and EST) to reduce the overheads needed for the method NAIV. The first method (RSY M) takes notice of symmetric properties on the system CS/sub N/ and considers the case where a single transient failure occurs on a specific server site. The second method (EST) verifies the responsiveness on the restricted system CS/sub 2/, and then estimates from its result the responsiveness on the system CS/sub N/. Then we perform simulation experiments to evaluate the proposed methods. The experimental results show that: the first method RSY M can reduce the overheads of the method NAIV, while keeping accuracy in evaluating the recovery time; and the second method EST succeeds in both reducing further the overheads of the method RSY M and assuring good approximation for the recovery time.