R-tree is widely used in indexing of multidimensional and spatial data.When it is used in environment of COW(Cluster Of Workstations), to the best of our knowledge it is used in Master-Slave mode or its variants. Under this mode when an entry is accessed, it must be locked. The higher the locked entry in the R-tree, the more transactions will be stopped from accessing all the offspring of it. Therefore the degree of concurrency is paralyzed. Moreover, the Master will become the hotspot in parallel processing which can worsen the overall performance. In this paper we present an upgraded parallel R-tree model which is suitable for the environment of COW. Our parallel R-tree can not only balance the load among processors but also decrease the conflicts of intraand inter-transactions. We can successfully prevent the emergence of hot spot from accessing the R-tree. The detailed searching algorithm and the load shipping algorithm of this parallel R-tree are given . Lastly we provide experimental results which show that the parallel R-tree performs well as expected.