Most real-time systems keep the whole databases in main memory for high performance. In this paper we presuppose that the system has client-server architecture with the restrictions on the memory and the time. In the light of these presupposition, we suggest an effective recovery technique for the index structures, a part of the database, and describe how it provides further performance improvement in the real-time client-server system. The basic paradigm of the proposed recovery technique is that the recovery technique of the indices should be designed differently from that of the other data part to tackle the constraints of the systems. The more client systems are appended to the system, the better performance can be guaranteed in the proposed recovery technique compared with the previous techniques.