Event notification service (ENS) system is a data dissemination technique which asynchronously notifies consumers whose interests match with the events published by producers. In a wide-area network, it is expected that some or other link/node failures to occur on a regular basis. Therefore, fault-tolerance is critical for smooth operation of a large-scale event notification service system. In this paper, we first describe architecture of event notification service system for large scale network to minimize the size of routing tables and to reduce latency of notification delivery to consumers. We then present a replication algorithm based on primary-backup replication so that event notification service system is resilient to failures of event servers and links between them, and ensures dissemination of events. The replication technique used in proposed architecture can minimize the impact of faults occurring in ENS system.