Reports

A File System Component Compiler

Zadok, Erez

File System development is a difficult and time consuming task, the results of which are rarely portable across operating systems. Several proposals to improve the vnode interface to allow for more flexible file system design and implementation have been made in recent years, but none is used in practice because they require costly fundamental changes to kernel interfaces, only operating systems vendors can make those changes, are still non-portable, tend to degrade performance, and do not appear to provide immediate return on such an investment. This proposal advocates a language for describing file systems, called FiST. The associated translator can generate portable C code — kernel resident or not — that implements the described file system. No kernel source code is needed and no existing vnode interface must change. The performance of the file systems automatically generated by FiST can be within a few percent of comparable hand-written file systems. The main benefits to automation are that development and maintenance costs are greatly reduced, and that it becomes practical to prototype, implement, test, debug, and compose a vastly larger set of such file systems with different properties. The proposed thesis will describe the language and its translator, use it to implement a few file systems on more than one platform, and evaluate the performance of the automatically generated code.

Subjects

Files

More About This Work

Academic Units
Computer Science
Publisher
Department of Computer Science, Columbia University
Series
Columbia University Computer Science Technical Reports, CUCS-033-97
Published Here
April 25, 2011