Context data is important information for catching the behaviors of applications in a pervasive space. To effectively store huge amount of data, tree-like layered storage architecture is proposed, where the leaf nodes collect data from sensing devices. In order to integrate data from mobile devices, the related leaf nodes that get data from the same device should upload and store the data to the host node. This paper presents a deep study of the data storage problem and proposes a global algorithm GHS and an online algorithm DHS to dynamically select the host node, which reduces the communication cost significantly. This paper also gives the theoretical and experimental analysis of these algorithms, which shows both GHS and DHS are correct and effective.