Data replication is a general mechanism to improve performance and availability for data grid applications.Distributing data replica reasonably in large scale data grid systems can decrease communication cost and improve its performances.This paper proposes a new Calking Dynamic Replication Distribution Algorithm (CDRDA) considering the communication cost among storage resource brokers and the characteristic of data accessing.By the algorithm

data replica will be distributed as a multi-level virtual cache and saved at the node in which applications can acquires the average lower communication cost.At last

the performance of the algorithm and its applications are also introduced.