Replica voting: a distributed middleware service for real-time dependable systems

In Information Assurance settings, majority voting among replica processes enhances the trust-worthiness of data collected from a hostile external environment. It allows a correct data fusion and dissemination by the end-users, in the presence of content corruptions and/or timing failures that may possibly occur during data collection. Two key elements are required of the voting functionality: i) fielding asynchronously generated real-time or near-real-time data, and ii) handling large sized non-numeric data sets with problemspecific interpretations - e.g., terrain images from radar stations. Under the constraints (i) and (ii), we describe a highly asynchronous voting service to effect a timely and low-overhead delivery of data to the users. The paper analyzes the service-level properties of a voting machinery to meet the stringent needs of IA applications. The externally visible properties are prescribed in terms of 'safety' and 'liveness' requirements of the underlying voting protocols that reflect the application-specific data integrity and availability constraints. When voting is provided as a middleware service, an application designer may prescribe the necessary service-level parameters: the level of resiliency and the performance aspects of data delivery.