Active File Systems for Data Mining and Multimedia

Abstract

Data mining and multimedia applications require huge amounts of storage. These applications are also compute-intensive. Active disks make use of the computational power available in the disk to reduce storage traffic. Many of the file system proposals for active disks work at the block level. In this paper we argue for the necessity of filtering at application level. We propose two file systems for active disks: active file system (ACFS) which binds files and filters at the file system level and active network file system (ANFS) which extends ACFS over networks. These file systems preserve the familiar Unix file system semantics to a large extent. We present an implementation of the file systems which makes minimal changes to the existing file system code in Linux.