Server placement optimization is one of the important contents of online media services system construction. How the system spends the least deployment cost to provide users with sufficient bandwidth is the optimization goal of the server placement problem. The problem is a NP-hard combinatorial optimization problem. Evolutionary algorithms show outstanding global optimization capabilities in this type of optimization problem compared to constructive heuristic method. However, the high dimensionality of the problem and the complicated evaluation of the deployment plan make the evolutionary algorithm easy to fall into the local optimum, and the computational cost is high. In this research, we propose an evolutionary framework for the server layout optimization, which can greatly improve the optimization efficiency of evolutionary algorithms and reduce the computational cost of the algorithm. In the framework, an offline-learning based approach is used to reduce the search space and a self-examining guided local search method is proposed to improve the search efficiency. Moreover, a look-up table based hybrid approach use to solution evaluation which can reduce computational overhead. Experiment results show that the proposed framework and optimization strategy can greatly improve the evolutionary algorithm search efficiency and has a good convergence performance.