Active disk file system: A distributed, scalable file system

Consistent improvements in processor and memory technology have led to disks having greater processing power and cache memory in a compact size. This increased processing power and memory allows disks to execute more than just the basic disk operations, sometimes, even run user defined codes. Offloading part of the application processing to the disks can help reducing the latency of data manipulation as well as the amount of data transferred across the network. Such disks are called active disks. In this paper, a file system for the active-disk-based data server (ADFS) is proposed. All data files stored on active disks are provided with operations, forming objects. For some applications such as database, application-specific operations can be run by disk processors so that only the results are returned to clients, rather than whole data file is read by the clients. Therefore, ADFS is able to reduce the application-processing overhead of the system. In ADFS, part of file system functionality of the file manager can be delegated to active disks. Our implementation is based on CORBA specification. A set of file system interfaces is implemented for each module of each file system component. Part of the file system functionality, such as lookup, is embedded in the modules for active disks. We tested the performance of ADFS with 64 files in three-layer directory system. The performance results are obtained in a limited manner and presented in this paper. Our ADFS is basically stateful, but the delegation of file system functionality to disks allows our file system to be more scalable and helps to overcome some of the limitations of current prevailing file systems, like NFS, by greatly reducing the work of the central file manager. In addition, it is conceivable that the central file manager could be eliminated altogether.