With the ever-increasing development of Internet, today抯 data-centric applications require storage platforms not only to possess some properties including high capacity and high scalability, but also to support structured data directly, which is beyond the abilities of current file systems and DBMS. In this paper, a model of network-attached object-based storage devices is presented. It provides storage and index interfaces for structured data employing the power of embedded processors, which eliminates the bottleneck of conventional storage systems. Then the architecture of cluster storage based on the model OStorage is proposed. It introduces the uniform storage of data and meta-data as well as query-locating mechanism to adapt storage hierarchies to structured data. These two methods improve the scalability and availability of the storage system. The prototype of Ostorage is implemented. The experimental results show that the throughput of it increases with the system scale linearly.