An object interface storage node for clustered file systems

In order to sustain scalability, clustered file systems distribute files across multiple I/O nodes in the cluster. The basis of many of these file systems is an object storage architecture where data is represented by variable sized containers called objects. The I/O nodes, however, use a POSIX file and directory interface which do not map well to the object interface. As a result, object storage nodes still have to resolve file paths and perform expensive lookup operations to find inodes and open objects stored as files. In this paper, we propose an alternative object based data interface where I/O node data can be accessed directly using object IDs instead of character strings. We modified an existing Linux file system to provide the object based interface which allows for faster data access times compared to the traditional directory-based interface. In addition, we modified a Object Storage Device (OSD) to use our object based interface to gain improvements in object creation and access. We believe this object based interface provides a useful alternative to the existing interface to provide better performance for cluster storage systems.