Average message overhead of replica control protocols

Abstract

Management of replicated data has received considerable attention in the last few years. Several replica control schemes have been proposed which work in the presence of both node and communication link failures. However, this resiliency to failure inflicts a performance penalty in terms of the communication overhead incurred. Though the issue of performance of these schemes, from the standpoint of availability of the system, has been well addressed, the issue of message overhead has been limited to the analysis of worst-case and best-case message bounds. In this paper, we compare several well-known replica management protocols and control schemes in terms of their average-case message overhead. We also consider the tradeoff between the message overhead and availability, and we define the system model considered. Analytical expressions are derived for five well-known replica control protocols. The results are discussed with numerical examples.<>