Since implicit restarted Arnoldi algorithm cannot obtain all eigenvalues in a designated area automatically, an eigenvalue searching strategy based on cluster computing is proposed to satisfy the requirements of accuracy and speed of online small signal stability analysis for ultra large scale power grid. Searching area is divided into several areas with same size and shift is appointed in each area to determine areas consisting more eigenvalues. Such area will be designated multi shifts to improve calculating speed. Square will be chosen to divide the searching area. Uncovered square can be identified automatically by comparing with the inscribed square of the circle area formed by the eigenvalues of a shift. The analysis results of East China Power Grid online data validate the correctness and efficiency of the proposed strategy. This work is supported by China Postdoctoral Science Foundation (No. 2016M591738) and Postdoctoral Science and Technology Program of Jiangsu Province (No. 1601160C).