Distributed storage system is an important research area in peer-to-peer technology. Current research on p2p structure has made a highly controlled routing scheme with limited hops of message transfer. People now turn to pursue lower network latency that is more factual. As a storage application, distributed system must have fault tolerance-recovery capability. Based on the analysis of current research, a computing modal more approximate to real time network is constructed. A computed shortest path of nodes is used to dynamically estimate the actual latency. Adjacent nodes are gathered under an evaluating algorithm to make node latency in the same group minimal. Thus a more efficient routing can be based on node grouping. For storage persistency, an interaction management and corresponded data transfer algorithm is presented. Its locality greatly enhances the system抯 response to all kinds of events, and ensures the system抯 availability. The simulation results are provided to show that the introduction of grouping truly helps to get an effective judgment on routing choice, and can be extend to a larger scale.