Abstract: Mining maximum frequent itemsets is a key problem in data mining field with numerous important applications. The present algorithms need scanning the database many times for updating the set of maximum frequent itemsets and are based on local databases. The algorithms of mining global maximum frequent itemsets are very few. Therefore, an algorithm for mining global maximum frequent itemsets is proposed, which can conveniently get all global maximum frequent itemsets using FP-tree structure by one time mining, and superset checking is very simple and speedy. FP-tree structure has provided a kind of convenient depth-first mining method. The algorithm combines FP-tree with restrained sub-tree for mining global maximum frequent itemsets and adopts an efficient distributed PDDM algorithm for broadcasting itemsets information and improves the expansibility and the concurrence. The PDDM algorithm is based on previous DDM algorithm and improves I?O problem and communication of previous distributed algorithms. Experimental results testify the feasibility and effectiveness of the algorithm.