An Adaptive Version-Controlled File System

1
An Adaptive Version-Controlled File System

• Makoto Matsushita, Tetsuo Yamamoto
• and Katsuro Inoue
• Osaka University, JAPAN

2
Background

• A research area of Version Management
• for successful software development project
• manages materials which is created by process
• In the coding process...
• version control model for source code, and
• its implementation by
• tools a special commands for management
• file systems a special structure within
  environments

3
Previous approaches (1)

• RCS (Revision Control System)
• RCS file and associated tools
• using diffs to save contents of versions
• using check-in/check-out command explicitly
• Engineers must learn how to use RCS tools
• Miss-operation of tools causes some unwanted
  result (file corruptions, etc).

4
Previous approaches (2)

• 3D File system
• management tools are integrated to software
  development environment itself
• each version is identified as filename version
• It employs its original file formats saved
  versions can be used by its own system.
• Version control models are statically bound to
  its implementation.

5
Our research objective

• Establish a new version control system
• more user-friendly and easy-to-use
• independent from various version management model
• An adaptive version-controlled file system
• auto check-in/check-out by file system
• employing external version control system

6
VCFS

• VCFS is composed of file system (VFS), version
  control daemon (VCD), version management
  sub-system, and VCFS control commands.

7
VCFS features

• Ordinal file operations such as read and
  write from/to files are treated as operations
  of version management.
• newly created files are treated as entry a new
  files to management system.
• modified existing files are treated as entry an
  another version of files to management system.
• reading files are treated as extract recent
  version of files from management system.

8
Stackable file system and VFS

• VFS is designed as stackable file system which
  stays on upper layer of other file system.
• VFS does not change the structure of lower layer
  file system
• Files by VFS are ordinal files on lower file
  system (such as UFS).

9
Version management subsystems

• VCFS simply employs existing version management
  system as version management subsystem.
• Version control daemon manages subsystems
• Users can select their own subsystem
• Two ore more subsystem can be used at the same
  time

10
VCFS prototype

• Current prototype of VCFS runs on UNIX
  environment (FreeBSD 3.0-RELEASE).
• VFS is implemented as kernel module.
• Using UFS (UNIX File System, a.k.a. FFS) as lower
  layer of file system.
• Total 5000 LOCs.

11
VCFS evaluation

• The purpose of evaluation
• VCFS does not kill system performance impatient
  engineer does not want slow software development
  environment.
• Evaluation overview
• Compared UFS, NULLFS, and VFS.
• Elapsed time of file read/write and application
  build, and total file size are measured.

12
Evaluation (read)

• We measured an elapsed time to read 1Mbytes of
  several files.

13
Evaluation (write)

• We measured an elapsed time to write 1Mbytes of
  several files.

14
Evaluation (application build)
15
Evaluation (file size)

• We applied a sample development data from a
  seminar of our university.

16
Evaluation summary

• VCFS is about 20 slower than UFS, but it is
  totally acceptable for practical use.
• Total file size of VCFS and RCS subsystem
  combination requires reasonable disk space.

17
Conclusion

• We propose VCFS, an adaptive version-controlled
  file system.
• Stackable VFS acts an intelligent wrapper for
  current version management system.
• Engineers can introduce their own subsystem.
• Evaluation results shows VCFS is usable in actual
  software development.

18
Further works

• Managing histories for directory, file owner, etc
  that are not scope of current system
• Of course, a new subsystem to manage such
  information will help.
• System support for existing subsystem is also
  needed.

19
Further works (cont.)

• Supporting distributed environment (data
  replication, etc)
• Date replication technology
• Cooperation with distributed file system.
• Using VCFS as data collection system for process
  metrics
• Visualization tool for VCFS data