In a distributed real time database system (DRTDBS), replication technique is used to easily meet the timeliness demand of real time transactions. The features of replication technique depend on the Replication Protocols (RPLs). Existing RPLs are kernel-based which is tightly linked with the concurrency control protocol and confined to a vendor-specific database. Middleware is the evolution of the kernel-based RPL that shifts the load of maintaining mutual consistency from the database kernel to the middleware server. Middleware based RPLs suffer from the issue of QoS, security-constraint, and dependency relationship. Here, our objective is to maintain the QoS during overload occurrence in the replicated DRTDBS (RDRTDBS). The proposed framework in this paper consists of a system model for expressing QoS requirements, a middleware extended with the prediction technique to predict the RTT processing time, and a set of policies to guarantee QoS. The performance of our proposed algorithms is compared with other baseline approaches and simulation results show that it outperforms a set of existing algorithms that use feedback control to overcome overload.