After analyzing the characteristics of overlay network, a hybrid K-medoids genetic model (HKGM) is proposed based on the hybrid clustering algorithms, which is used to choose the multicast service nodes (MSNs) from network nodes. Compared with the traditional K-medoids model, HKGM not only avoids converging to local minimum value, but also is robust to initialization. Also, during the evolution, according to actual features of MSNs in overlay network, the evolutional control strategies including diverse gene and evolutionary elite reservation are used to enhance the local search ability of model, and to increase the convergent speed.