Based on the real-time features of transactions and data, an intrusion tolerant architecture is proposed for the real-time database system. Considering real-time parameters and intrusion detection latency, Semi-Markov evaluation model is established to access the survivability of real-time database. Based on this model, relevant quantitative criteria are given to define the indicators of the survivability, such as integrity and availability, so as to validate the survivability of real-time database. Because the factors such as false alarm, detection rate and the intensity of attacks have important effects on the survivability, they are analyzed by the TPC-C benchmark in detail. Experiments show that the model can accurately predict the behaviors of real-time database. The proposed real-time database with intrusion tolerance can still keep the basic survival when facing attacks.