Based on Chip Multi-Processor(CMP), this paper presents optimization of cluster join in Radix-Join algorithm. In order to solve the problem of serious cache may miss during the multithreaded cluster join execution, and to improve the performance of cache access, we adopt preload thread to read the clusters whose thread will access from memory to L2-Cache. Furthermore, based on the cost model of cluster join execution, the framework of cluster join execution and various thread parameters have been optimized. In the experiments, we implement the algorithm in EaseDB. The results show that cluster join performance is improved.